Datasets:

CShorten

/

Last-Week-on-ML-ArXiv

like 0

Less Is More: Fast Multivariate Time Series Forecasting with Light Sampling-oriented MLP Structures

Multivariate time series forecasting has seen widely ranging applications in various domains, including finance, traffic, energy, and healthcare. To capture the sophisticated temporal patterns, plenty of research studies designed complex neural network architectures based on many variants of RNNs, GNNs, and Transformers. However, complex models are often computationally expensive and thus face a severe challenge in training and inference efficiency when applied to large-scale real-world datasets. In this paper, we introduce LightTS, a light deep learning architecture merely based on simple MLP-based structures. The key idea of LightTS is to apply an MLP-based structure on top of two delicate down-sampling strategies, including interval sampling and continuous sampling, inspired by a crucial fact that down-sampling time series often preserves the majority of its information. We conduct extensive experiments on eight widely used benchmark datasets. Compared with the existing state-of-the-art methods, LightTS demonstrates better performance on five of them and comparable performance on the rest. Moreover, LightTS is highly efficient. It uses less than 5% FLOPS compared with previous SOTA methods on the largest benchmark dataset. In addition, LightTS is robust and has a much smaller variance in forecasting accuracy than previous SOTA methods in long sequence forecasting tasks.

An Improved Probability Propagation Algorithm for Density Peak Clustering Based on Natural Nearest Neighborhood

Clustering by fast search and find of density peaks (DPC) (Since, 2014) has been proven to be a promising clustering approach that efficiently discovers the centers of clusters by finding the density peaks. The accuracy of DPC depends on the cutoff distance ($d_c$), the cluster number ($k$) and the selection of the centers of clusters. Moreover, the final allocation strategy is sensitive and has poor fault tolerance. The shortcomings above make the algorithm sensitive to parameters and only applicable for some specific datasets. To overcome the limitations of DPC, this paper presents an improved probability propagation algorithm for density peak clustering based on the natural nearest neighborhood (DPC-PPNNN). By introducing the idea of natural nearest neighborhood and probability propagation, DPC-PPNNN realizes the nonparametric clustering process and makes the algorithm applicable for more complex datasets. In experiments on several datasets, DPC-PPNNN is shown to outperform DPC, K-means and DBSCAN.

Portuguese Man-of-War Image Classification with Convolutional Neural Networks

Portuguese man-of-war (PMW) is a gelatinous organism with long tentacles capable of causing severe burns, thus leading to negative impacts on human activities, such as tourism and fishing. There is a lack of information about the spatio-temporal dynamics of this species. Therefore, the use of alternative methods for collecting data can contribute to their monitoring. Given the widespread use of social networks and the eye-catching look of PMW, Instagram posts can be a promising data source for monitoring. The first task to follow this approach is to identify posts that refer to PMW. This paper reports on the use of convolutional neural networks for PMW images classification, in order to automate the recognition of Instagram posts. We created a suitable dataset, and trained three different neural networks: VGG-16, ResNet50, and InceptionV3, with and without a pre-trained step with the ImageNet dataset. We analyzed their results using accuracy, precision, recall, and F1 score metrics. The pre-trained ResNet50 network presented the best results, obtaining 94% of accuracy and 95% of precision, recall, and F1 score. These results show that convolutional neural networks can be very effective for recognizing PMW images from the Instagram social media.

How Robust is Your Fairness? Evaluating and Sustaining Fairness under Unseen Distribution Shifts

Increasing concerns have been raised on deep learning fairness in recent years. Existing fairness-aware machine learning methods mainly focus on the fairness of in-distribution data. However, in real-world applications, it is common to have distribution shift between the training and test data. In this paper, we first show that the fairness achieved by existing methods can be easily broken by slight distribution shifts. To solve this problem, we propose a novel fairness learning method termed CUrvature MAtching (CUMA), which can achieve robust fairness generalizable to unseen domains with unknown distributional shifts. Specifically, CUMA enforces the model to have similar generalization ability on the majority and minority groups, by matching the loss curvature distributions of the two groups. We evaluate our method on three popular fairness datasets. Compared with existing methods, CUMA achieves superior fairness under unseen distribution shifts, without sacrificing either the overall accuracy or the in-distribution fairness.

Partial and Asymmetric Contrastive Learning for Out-of-Distribution Detection in Long-Tailed Recognition

Existing out-of-distribution (OOD) detection methods are typically benchmarked on training sets with balanced class distributions. However, in real-world applications, it is common for the training sets to have long-tailed distributions. In this work, we first demonstrate that existing OOD detection methods commonly suffer from significant performance degradation when the training set is long-tail distributed. Through analysis, we posit that this is because the models struggle to distinguish the minority tail-class in-distribution samples, from the true OOD samples, making the tail classes more prone to be falsely detected as OOD. To solve this problem, we propose Partial and Asymmetric Supervised Contrastive Learning (PASCL), which explicitly encourages the model to distinguish between tail-class in-distribution samples and OOD samples. To further boost in-distribution classification accuracy, we propose Auxiliary Branch Finetuning, which uses two separate branches of BN and classification layers for anomaly detection and in-distribution classification, respectively. The intuition is that in-distribution and OOD anomaly data have different underlying distributions. Our method outperforms previous state-of-the-art method by $1.29\%$, $1.45\%$, $0.69\%$ anomaly detection false positive rate (FPR) and $3.24\%$, $4.06\%$, $7.89\%$ in-distribution classification accuracy on CIFAR10-LT, CIFAR100-LT, and ImageNet-LT, respectively. Code and pre-trained models are available at https://github.com/amazon-research/long-tailed-ood-detection.

Removing Batch Normalization Boosts Adversarial Training

Adversarial training (AT) defends deep neural networks against adversarial attacks. One challenge that limits its practical application is the performance degradation on clean samples. A major bottleneck identified by previous works is the widely used batch normalization (BN), which struggles to model the different statistics of clean and adversarial training samples in AT. Although the dominant approach is to extend BN to capture this mixture of distribution, we propose to completely eliminate this bottleneck by removing all BN layers in AT. Our normalizer-free robust training (NoFrost) method extends recent advances in normalizer-free networks to AT for its unexplored advantage on handling the mixture distribution challenge. We show that NoFrost achieves adversarial robustness with only a minor sacrifice on clean sample accuracy. On ImageNet with ResNet50, NoFrost achieves $74.06\%$ clean accuracy, which drops merely $2.00\%$ from standard training. In contrast, BN-based AT obtains $59.28\%$ clean accuracy, suffering a significant $16.78\%$ drop from standard training. In addition, NoFrost achieves a $23.56\%$ adversarial robustness against PGD attack, which improves the $13.57\%$ robustness in BN-based AT. We observe better model smoothness and larger decision margins from NoFrost, which make the models less sensitive to input perturbations and thus more robust. Moreover, when incorporating more data augmentations into NoFrost, it achieves comprehensive robustness against multiple distribution shifts. Code and pre-trained models are public at https://github.com/amazon-research/normalizer-free-robust-training.

It's all About Consistency: A Study on Memory Composition for Replay-Based Methods in Continual Learning

Continual Learning methods strive to mitigate Catastrophic Forgetting (CF), where knowledge from previously learned tasks is lost when learning a new one. Among those algorithms, some maintain a subset of samples from previous tasks when training. These samples are referred to as a memory. These methods have shown outstanding performance while being conceptually simple and easy to implement. Yet, despite their popularity, little has been done to understand which elements to be included into the memory. Currently, this memory is often filled via random sampling with no guiding principles that may aid in retaining previous knowledge. In this work, we propose a criterion based on the learning consistency of a sample called Consistency AWare Sampling (CAWS). This criterion prioritizes samples that are easier to learn by deep networks. We perform studies on three different memory-based methods: AGEM, GDumb, and Experience Replay, on MNIST, CIFAR-10 and CIFAR-100 datasets. We show that using the most consistent elements yields performance gains when constrained by a compute budget; when under no such constrain, random sampling is a strong baseline. However, using CAWS on Experience Replay yields improved performance over the random baseline. Finally, we show that CAWS achieves similar results to a popular memory selection method while requiring significantly less computational resources.

DecisioNet -- A Binary-Tree Structured Neural Network

Deep neural networks (DNNs) and decision trees (DTs) are both state-of-the-art classifiers. DNNs perform well due to their representational learning capabilities, while DTs are computationally efficient as they perform inference along one route (root-to-leaf) that is dependent on the input data. In this paper, we present DecisioNet (DN), a binary-tree structured neural network. We propose a systematic way to convert an existing DNN into a DN to create a lightweight version of the original model. DecisioNet takes the best of both worlds - it uses neural modules to perform representational learning and utilizes its tree structure to perform only a portion of the computations. We evaluate various DN architectures, along with their corresponding baseline models on the FashionMNIST, CIFAR10, and CIFAR100 datasets. We show that the DN variants achieve similar accuracy while significantly reducing the computational cost of the original network.

Saliency-Regularized Deep Multi-Task Learning

Multitask learning is a framework that enforces multiple learning tasks to share knowledge to improve their generalization abilities. While shallow multitask learning can learn task relations, it can only handle predefined features. Modern deep multitask learning can jointly learn latent features and task sharing, but they are obscure in task relation. Also, they predefine which layers and neurons should share across tasks and cannot learn adaptively. To address these challenges, this paper proposes a new multitask learning framework that jointly learns latent features and explicit task relations by complementing the strength of existing shallow and deep multitask learning scenarios. Specifically, we propose to model the task relation as the similarity between task input gradients, with a theoretical analysis of their equivalency. In addition, we innovatively propose a multitask learning objective that explicitly learns task relations by a new regularizer. Theoretical analysis shows that the generalizability error has been reduced thanks to the proposed regularizer. Extensive experiments on several multitask learning and image classification benchmarks demonstrate the proposed method effectiveness, efficiency as well as reasonableness in the learned task relation patterns.

USHER: Unbiased Sampling for Hindsight Experience Replay

Dealing with sparse rewards is a long-standing challenge in reinforcement learning (RL). Hindsight Experience Replay (HER) addresses this problem by reusing failed trajectories for one goal as successful trajectories for another. This allows for both a minimum density of reward and for generalization across multiple goals. However, this strategy is known to result in a biased value function, as the update rule underestimates the likelihood of bad outcomes in a stochastic environment. We propose an asymptotically unbiased importance-sampling-based algorithm to address this problem without sacrificing performance on deterministic environments. We show its effectiveness on a range of robotic systems, including challenging high dimensional stochastic environments.

Learning Noise with Generative Adversarial Networks: Explorations with Classical Random Process Models

Random noise arising from physical processes is an inherent characteristic of measurements and a limiting factor for most signal processing tasks. Given the recent interest in generative adversarial networks (GANs) for data-driven signal modeling, it is important to determine to what extent GANs can faithfully reproduce noise in target data sets. In this paper, we present an empirical investigation that aims to shed light on this issue for time series. Namely, we examine the ability of two general-purpose time-series GANs, a direct time-series model and an image-based model using a short-time Fourier transform (STFT) representation, to learn a broad range of noise types commonly encountered in electronics and communication systems: band-limited thermal noise, power law noise, shot noise, and impulsive noise. We find that GANs are capable of learning many noise types, although they predictably struggle when the GAN architecture is not well suited to some aspects of the noise, e.g., impulsive time-series with extreme outliers. Our findings provide insights into the capabilities and potential limitations of current approaches to time-series GANs and highlight areas for further research. In addition, our battery of tests provides a useful benchmark to aid the development of deep generative models for time series.

Mathematical Foundations of Graph-Based Bayesian Semi-Supervised Learning

In recent decades, science and engineering have been revolutionized by a momentous growth in the amount of available data. However, despite the unprecedented ease with which data are now collected and stored, labeling data by supplementing each feature with an informative tag remains to be challenging. Illustrative tasks where the labeling process requires expert knowledge or is tedious and time-consuming include labeling X-rays with a diagnosis, protein sequences with a protein type, texts by their topic, tweets by their sentiment, or videos by their genre. In these and numerous other examples, only a few features may be manually labeled due to cost and time constraints. How can we best propagate label information from a small number of expensive labeled features to a vast number of unlabeled ones? This is the question addressed by semi-supervised learning (SSL). This article overviews recent foundational developments on graph-based Bayesian SSL, a probabilistic framework for label propagation using similarities between features. SSL is an active research area and a thorough review of the extant literature is beyond the scope of this article. Our focus will be on topics drawn from our own research that illustrate the wide range of mathematical tools and ideas that underlie the rigorous study of the statistical accuracy and computational efficiency of graph-based Bayesian SSL.

Folding over Neural Networks

Neural networks are typically represented as data structures that are traversed either through iteration or by manual chaining of method calls. However, a deeper analysis reveals that structured recursion can be used instead, so that traversal is directed by the structure of the network itself. This paper shows how such an approach can be realised in Haskell, by encoding neural networks as recursive data types, and then their training as recursion scheme patterns. In turn, we promote a coherent implementation of neural networks that delineates between their structure and semantics, allowing for compositionality in both how they are built and how they are trained.

FasterAI: A Lightweight Library for Creating Sparse Neural Networks

FasterAI is a PyTorch-based library, aiming to facilitate the utilization of deep neural networks compression techniques such as sparsification, pruning, knowledge distillation, or regularization. The library is built with the purpose of enabling quick implementation and experimentation. More particularly, compression techniques are leveraging Callback systems of libraries such as fastai and Pytorch Lightning to bring a user-friendly and high-level API. The main asset of FasterAI is its lightweight, yet powerful, simplicity of use. Indeed, because it was developed in a very granular way, users can create thousands of unique experiments by using different combinations of parameters. In this paper, we focus on the sparsifying capabilities of FasterAI, which represents the core of the library. Performing sparsification of a neural network in FasterAI only requires a single additional line of code in the traditional training loop, yet allows to perform state-of-the-art techniques such as Lottery Ticket Hypothesis experiments

Patient-specific modelling, simulation and real time processing for constrictive respiratory diseases

Asthma is a common chronic disease of the respiratory system causing significant disability and societal burden. It affects over 500 million people worldwide and generates costs exceeding $USD 56 billion in 2011 in the United States. Managing asthma involves controlling symptoms, preventing exacerbations, and maintaining lung function. Improving asthma control affects the daily life of patients and is associated with a reduced risk of exacerbations and lung function impairment, reduces the cost of asthma care and indirect costs associated with reduced productivity. Understanding the complex dynamics of the pulmonary system and the lung's response to disease, injury, and treatment is fundamental to the advancement of Asthma treatment. Computational models of the respiratory system seek to provide a theoretical framework to understand the interaction between structure and function. Their application can improve pulmonary medicine by a patient-specific approach to medicinal methodologies optimizing the delivery given the personalized geometry and personalized ventilation patterns while introducing a patient-specific technique that maximizes drug delivery. A three-fold objective addressed within this dissertation becomes prominent at this point. The first part refers to the comprehension of pulmonary pathophysiology and the mechanics of Asthma and subsequently of constrictive pulmonary conditions in general. The second part refers to the design and implementation of tools that facilitate personalized medicine to improve delivery and effectiveness. Finally, the third part refers to the self-management of the condition, meaning that medical personnel and patients have access to tools and methods that allow the first party to easily track the course of the condition and the second party, i.e. the patient to easily self-manage it alleviating the significant burden from the health system.

NP-Match: When Neural Processes meet Semi-Supervised Learning

Semi-supervised learning (SSL) has been widely explored in recent years, and it is an effective way of leveraging unlabeled data to reduce the reliance on labeled data. In this work, we adjust neural processes (NPs) to the semi-supervised image classification task, resulting in a new method named NP-Match. NP-Match is suited to this task for two reasons. Firstly, NP-Match implicitly compares data points when making predictions, and as a result, the prediction of each unlabeled data point is affected by the labeled data points that are similar to it, which improves the quality of pseudo-labels. Secondly, NP-Match is able to estimate uncertainty that can be used as a tool for selecting unlabeled samples with reliable pseudo-labels. Compared with uncertainty-based SSL methods implemented with Monte Carlo (MC) dropout, NP-Match estimates uncertainty with much less computational overhead, which can save time at both the training and the testing phases. We conducted extensive experiments on four public datasets, and NP-Match outperforms state-of-the-art (SOTA) results or achieves competitive results on them, which shows the effectiveness of NP-Match and its potential for SSL.

Distributed Online System Identification for LTI Systems Using Reverse Experience Replay

Identification of linear time-invariant (LTI) systems plays an important role in control and reinforcement learning. Both asymptotic and finite-time offline system identification are well-studied in the literature. For online system identification, the idea of stochastic-gradient descent with reverse experience replay (SGD-RER) was recently proposed, where the data sequence is stored in several buffers and the stochastic-gradient descent (SGD) update performs backward in each buffer to break the time dependency between data points. Inspired by this work, we study distributed online system identification of LTI systems over a multi-agent network. We consider agents as identical LTI systems, and the network goal is to jointly estimate the system parameters by leveraging the communication between agents. We propose DSGD-RER, a distributed variant of the SGD-RER algorithm, and theoretically characterize the improvement of the estimation error with respect to the network size. Our numerical experiments certify the reduction of estimation error as the network size grows.

Identifying the Context Shift between Test Benchmarks and Production Data

Across a wide variety of domains, there exists a performance gap between machine learning models' accuracy on dataset benchmarks and real-world production data. Despite the careful design of static dataset benchmarks to represent the real-world, models often err when the data is out-of-distribution relative to the data the models have been trained on. We can directly measure and adjust for some aspects of distribution shift, but we cannot address sample selection bias, adversarial perturbations, and non-stationarity without knowing the data generation process. In this paper, we outline two methods for identifying changes in context that lead to distribution shifts and model prediction errors: leveraging human intuition and expert knowledge to identify first-order contexts and developing dynamic benchmarks based on desiderata for the data generation process. Furthermore, we present two case-studies to highlight the implicit assumptions underlying applied machine learning models that tend to lead to errors when attempting to generalize beyond test benchmark datasets. By paying close attention to the role of context in each prediction task, researchers can reduce context shift errors and increase generalization performance.

Protea: Client Profiling within Federated Systems using Flower

Federated Learning (FL) has emerged as a prospective solution that facilitates the training of a high-performing centralised model without compromising the privacy of users. While successful, research is currently limited by the possibility of establishing a realistic large-scale FL system at the early stages of experimentation. Simulation can help accelerate this process. To facilitate efficient scalable FL simulation of heterogeneous clients, we design and implement Protea, a flexible and lightweight client profiling component within federated systems using the FL framework Flower. It allows automatically collecting system-level statistics and estimating the resources needed for each client, thus running the simulation in a resource-aware fashion. The results show that our design successfully increases parallelism for 1.66 $\times$ faster wall-clock time and 2.6$\times$ better GPU utilisation, which enables large-scale experiments on heterogeneous clients.

Generating gender-ambiguous voices for privacy-preserving speech recognition

Our voice encodes a uniquely identifiable pattern which can be used to infer private attributes, such as gender or identity, that an individual might wish not to reveal when using a speech recognition service. To prevent attribute inference attacks alongside speech recognition tasks, we present a generative adversarial network, GenGAN, that synthesises voices that conceal the gender or identity of a speaker. The proposed network includes a generator with a U-Net architecture that learns to fool a discriminator. We condition the generator only on gender information and use an adversarial loss between signal distortion and privacy preservation. We show that GenGAN improves the trade-off between privacy and utility compared to privacy-preserving representation learning methods that consider gender information as a sensitive attribute to protect.

Learning to Increase the Power of Conditional Randomization Tests

The model-X conditional randomization test is a generic framework for conditional independence testing, unlocking new possibilities to discover features that are conditionally associated with a response of interest while controlling type-I error rates. An appealing advantage of this test is that it can work with any machine learning model to design powerful test statistics. In turn, the common practice in the model-X literature is to form a test statistic using machine learning models, trained to maximize predictive accuracy with the hope to attain a test with good power. However, the ideal goal here is to drive the model (during training) to maximize the power of the test, not merely the predictive accuracy. In this paper, we bridge this gap by introducing, for the first time, novel model-fitting schemes that are designed to explicitly improve the power of model-X tests. This is done by introducing a new cost function that aims at maximizing the test statistic used to measure violations of conditional independence. Using synthetic and real data sets, we demonstrate that the combination of our proposed loss function with various base predictive models (lasso, elastic net, and deep neural networks) consistently increases the number of correct discoveries obtained, while maintaining type-I error rates under control.

Comparative Analysis of Time Series Forecasting Approaches for Household Electricity Consumption Prediction

As a result of increasing population and globalization, the demand for energy has greatly risen. Therefore, accurate energy consumption forecasting has become an essential prerequisite for government planning, reducing power wastage and stable operation of the energy management system. In this work we present a comparative analysis of major machine learning models for time series forecasting of household energy consumption. Specifically, we use Weka, a data mining tool to first apply models on hourly and daily household energy consumption datasets available from Kaggle data science community. The models applied are: Multilayer Perceptron, K Nearest Neighbor regression, Support Vector Regression, Linear Regression, and Gaussian Processes. Secondly, we also implemented time series forecasting models, ARIMA and VAR, in python to forecast household energy consumption of selected South Korean households with and without weather data. Our results show that the best methods for the forecasting of energy consumption prediction are Support Vector Regression followed by Multilayer Perceptron and Gaussian Process Regression.

Recipe for Fast Large-scale SVM Training: Polishing, Parallelism, and more RAM!

Support vector machines (SVMs) are a standard method in the machine learning toolbox, in particular for tabular data. Non-linear kernel SVMs often deliver highly accurate predictors, however, at the cost of long training times. That problem is aggravated by the exponential growth of data volumes over time. It was tackled in the past mainly by two types of techniques: approximate solvers, and parallel GPU implementations. In this work, we combine both approaches to design an extremely fast dual SVM solver. We fully exploit the capabilities of modern compute servers: many-core architectures, multiple high-end GPUs, and large random access memory. On such a machine, we train a large-margin classifier on the ImageNet data set in 24 minutes.

Mental Illness Classification on Social Media Texts using Deep Learning and Transfer Learning

Given the current social distance restrictions across the world, most individuals now use social media as their major medium of communication. Millions of people suffering from mental diseases have been isolated due to this, and they are unable to get help in person. They have become more reliant on online venues to express themselves and seek advice on dealing with their mental disorders. According to the World health organization (WHO), approximately 450 million people are affected. Mental illnesses, such as depression, anxiety, etc., are immensely common and have affected an individuals' physical health. Recently Artificial Intelligence (AI) methods have been presented to help mental health providers, including psychiatrists and psychologists, in decision making based on patients' authentic information (e.g., medical records, behavioral data, social media utilization, etc.). AI innovations have demonstrated predominant execution in numerous real-world applications broadening from computer vision to healthcare. This study analyzes unstructured user data on the Reddit platform and classifies five common mental illnesses: depression, anxiety, bipolar disorder, ADHD, and PTSD. We trained traditional machine learning, deep learning, and transfer learning multi-class models to detect mental disorders of individuals. This effort will benefit the public health system by automating the detection process and informing appropriate authorities about people who require emergency assistance.

Facial Image Reconstruction from Functional Magnetic Resonance Imaging via GAN Inversion with Improved Attribute Consistency

Neuroscience studies have revealed that the brain encodes visual content and embeds information in neural activity. Recently, deep learning techniques have facilitated attempts to address visual reconstructions by mapping brain activity to image stimuli using generative adversarial networks (GANs). However, none of these studies have considered the semantic meaning of latent code in image space. Omitting semantic information could potentially limit the performance. In this study, we propose a new framework to reconstruct facial images from functional Magnetic Resonance Imaging (fMRI) data. With this framework, the GAN inversion is first applied to train an image encoder to extract latent codes in image space, which are then bridged to fMRI data using linear transformation. Following the attributes identified from fMRI data using an attribute classifier, the direction in which to manipulate attributes is decided and the attribute manipulator adjusts the latent code to improve the consistency between the seen image and the reconstructed image. Our experimental results suggest that the proposed framework accomplishes two goals: (1) reconstructing clear facial images from fMRI data and (2) maintaining the consistency of semantic characteristics.

Government Intervention in Catastrophe Insurance Markets: A Reinforcement Learning Approach

This paper designs a sequential repeated game of a micro-founded society with three types of agents: individuals, insurers, and a government. Nascent to economics literature, we use Reinforcement Learning (RL), closely related to multi-armed bandit problems, to learn the welfare impact of a set of proposed policy interventions per $1 spent on them. The paper rigorously discusses the desirability of the proposed interventions by comparing them against each other on a case-by-case basis. The paper provides a framework for algorithmic policy evaluation using calibrated theoretical models which can assist in feasibility studies.

Stabilizing Off-Policy Deep Reinforcement Learning from Pixels

Off-policy reinforcement learning (RL) from pixel observations is notoriously unstable. As a result, many successful algorithms must combine different domain-specific practices and auxiliary losses to learn meaningful behaviors in complex environments. In this work, we provide novel analysis demonstrating that these instabilities arise from performing temporal-difference learning with a convolutional encoder and low-magnitude rewards. We show that this new visual deadly triad causes unstable training and premature convergence to degenerate solutions, a phenomenon we name catastrophic self-overfitting. Based on our analysis, we propose A-LIX, a method providing adaptive regularization to the encoder's gradients that explicitly prevents the occurrence of catastrophic self-overfitting using a dual objective. By applying A-LIX, we significantly outperform the prior state-of-the-art on the DeepMind Control and Atari 100k benchmarks without any data augmentation or auxiliary losses.

SSD-Faster Net: A Hybrid Network for Industrial Defect Inspection

The quality of industrial components is critical to the production of special equipment such as robots. Defect inspection of these components is an efficient way to ensure quality. In this paper, we propose a hybrid network, SSD-Faster Net, for industrial defect inspection of rails, insulators, commutators etc. SSD-Faster Net is a two-stage network, including SSD for quickly locating defective blocks, and an improved Faster R-CNN for defect segmentation. For the former, we propose a novel slice localization mechanism to help SSD scan quickly. The second stage is based on improved Faster R-CNN, using FPN, deformable kernel(DK) to enhance representation ability. It fuses multi-scale information, and self-adapts the receptive field. We also propose a novel loss function and use ROI Align to improve accuracy. Experiments show that our SSD-Faster Net achieves an average accuracy of 84.03%, which is 13.42% higher than the nearest competitor based on Faster R-CNN, 4.14% better than GAN-based methods, more than 10% higher than that of DNN-based detectors. And the computing speed is improved by nearly 7%, which proves its robustness and superior performance.

Advancing protein language models with linguistics: a roadmap for improved interpretability

Deep neural-network-based language models (LMs) are increasingly applied to large-scale protein sequence data to predict protein function. However, being largely blackbox models and thus challenging to interpret, current protein LM approaches do not contribute to a fundamental understanding of sequence-function mappings, hindering rule-based biotherapeutic drug development. We argue that guidance drawn from linguistics, a field specialized in analytical rule extraction from natural language data, can aid with building more interpretable protein LMs that have learned relevant domain-specific rules. Differences between protein sequence data and linguistic sequence data require the integration of more domain-specific knowledge in protein LMs compared to natural language LMs. Here, we provide a linguistics-based roadmap for protein LM pipeline choices with regard to training data, tokenization, token embedding, sequence embedding, and model interpretation. Combining linguistics with protein LMs enables the development of next-generation interpretable machine learning models with the potential of uncovering the biological mechanisms underlying sequence-function relationships.

Renaissance Robot: Optimal Transport Policy Fusion for Learning Diverse Skills

Deep reinforcement learning (RL) is a promising approach to solving complex robotics problems. However, the process of learning through trial-and-error interactions is often highly time-consuming, despite recent advancements in RL algorithms. Additionally, the success of RL is critically dependent on how well the reward-shaping function suits the task, which is also time-consuming to design. As agents trained on a variety of robotics problems continue to proliferate, the ability to reuse their valuable learning for new domains becomes increasingly significant. In this paper, we propose a post-hoc technique for policy fusion using Optimal Transport theory as a robust means of consolidating the knowledge of multiple agents that have been trained on distinct scenarios. We further demonstrate that this provides an improved weights initialisation of the neural network policy for learning new tasks, requiring less time and computational resources than either retraining the parent policies or training a new policy from scratch. Ultimately, our results on diverse agents commonly used in deep RL show that specialised knowledge can be unified into a "Renaissance agent", allowing for quicker learning of new skills.

PrUE: Distilling Knowledge from Sparse Teacher Networks

Although deep neural networks have enjoyed remarkable success across a wide variety of tasks, their ever-increasing size also imposes significant overhead on deployment. To compress these models, knowledge distillation was proposed to transfer knowledge from a cumbersome (teacher) network into a lightweight (student) network. However, guidance from a teacher does not always improve the generalization of students, especially when the size gap between student and teacher is large. Previous works argued that it was due to the high certainty of the teacher, resulting in harder labels that were difficult to fit. To soften these labels, we present a pruning method termed Prediction Uncertainty Enlargement (PrUE) to simplify the teacher. Specifically, our method aims to decrease the teacher's certainty about data, thereby generating soft predictions for students. We empirically investigate the effectiveness of the proposed method with experiments on CIFAR-10/100, Tiny-ImageNet, and ImageNet. Results indicate that student networks trained with sparse teachers achieve better performance. Besides, our method allows researchers to distill knowledge from deeper networks to improve students further. Our code is made public at: \url{https://github.com/wangshaopu/prue}.

Features of a Splashing Drop on a Solid Surface and the Temporal Evolution extracted through Image-Sequence Classification using an Interpretable Feedforward Neural Network

This paper reports the features of a splashing drop on a solid surface and the temporal evolution, which are extracted through image-sequence classification using a highly interpretable feedforward neural network (FNN) with zero hidden layer. The image sequences used for training-validation and testing of the FNN show the early-stage deformation of milli-sized ethanol drops that impact a hydrophilic glass substrate with the Weber number ranges between 31-474 (splashing threshold about 173). Specific videographing conditions and digital image processing are performed to ensure the high similarity among the image sequences. As a result, the trained FNNs achieved a test accuracy higher than 96%. Remarkably, the feature extraction shows that the trained FNN identifies the temporal evolution of the ejected secondary droplets around the aerodynamically lifted lamella and the relatively high contour of the main body as the features of a splashing drop, while the relatively short and thick lamella as the feature of a nonsplashing drop. The physical interpretation for these features and their respective temporal evolution have been identified except for the difference in contour height of the main body between splashing and nonsplashing drops. The observation reported in this study is important for the development of a data-driven simulation for modeling the deformation of a splashing drop during the impact on a solid surface.

Task-Oriented Sensing, Computation, and Communication Integration for Multi-Device Edge AI

This paper studies a new multi-device edge artificial-intelligent (AI) system, which jointly exploits the AI model split inference and integrated sensing and communication (ISAC) to enable low-latency intelligent services at the network edge. In this system, multiple ISAC devices perform radar sensing to obtain multi-view data, and then offload the quantized version of extracted features to a centralized edge server, which conducts model inference based on the cascaded feature vectors. Under this setup and by considering classification tasks, we measure the inference accuracy by adopting an approximate but tractable metric, namely discriminant gain, which is defined as the distance of two classes in the Euclidean feature space under normalized covariance. To maximize the discriminant gain, we first quantify the influence of the sensing, computation, and communication processes on it with a derived closed-form expression. Then, an end-to-end task-oriented resource management approach is developed by integrating the three processes into a joint design. This integrated sensing, computation, and communication (ISCC) design approach, however, leads to a challenging non-convex optimization problem, due to the complicated form of discriminant gain and the device heterogeneity in terms of channel gain, quantization level, and generated feature subsets. Remarkably, the considered non-convex problem can be optimally solved based on the sum-of-ratios method. This gives the optimal ISCC scheme, that jointly determines the transmit power and time allocation at multiple devices for sensing and communication, as well as their quantization bits allocation for computation distortion control. By using human motions recognition as a concrete AI inference task, extensive experiments are conducted to verify the performance of our derived optimal ISCC scheme.

An Empirical Evaluation of $k$-Means Coresets

Coresets are among the most popular paradigms for summarizing data. In particular, there exist many high performance coresets for clustering problems such as $k$-means in both theory and practice. Curiously, there exists no work on comparing the quality of available $k$-means coresets. In this paper we perform such an evaluation. There currently is no algorithm known to measure the distortion of a candidate coreset. We provide some evidence as to why this might be computationally difficult. To complement this, we propose a benchmark for which we argue that computing coresets is challenging and which also allows us an easy (heuristic) evaluation of coresets. Using this benchmark and real-world data sets, we conduct an exhaustive evaluation of the most commonly used coreset algorithms from theory and practice.

Digital-twin-enhanced metal tube bending forming real-time prediction method based on Multi-source-input MTL

As one of the most widely used metal tube bending methods, the rotary draw bending (RDB) process enables reliable and high-precision metal tube bending forming (MTBF). The forming accuracy is seriously affected by the springback and other potential forming defects, of which the mechanism analysis is difficult to deal with. At the same time, the existing methods are mainly conducted in offline space, ignoring the real-time information in the physical world, which is unreliable and inefficient. To address this issue, a digital-twin-enhanced (DT-enhanced) metal tube bending forming real-time prediction method based on multi-source-input multi-task learning (MTL) is proposed. The new method can achieve comprehensive MTBF real-time prediction. By sharing the common feature of the multi-close domain and adopting group regularization strategy on feature sharing and accepting layers, the accuracy and efficiency of the multi-source-input MTL can be guaranteed. Enhanced by DT, the physical real-time deformation data is aligned in the image dimension by an improved Grammy Angle Field (GAF) conversion, realizing the reflection of the actual processing. Different from the traditional offline prediction methods, the new method integrates the virtual and physical data to achieve a more efficient and accurate real-time prediction result. and the DT mapping connection between virtual and physical systems can be achieved. To exclude the effects of equipment errors, the effectiveness of the proposed method is verified on the physical experiment-verified FE simulation scenarios. At the same time, the common pre-training networks are compared with the proposed method. The results show that the proposed DT-enhanced prediction method is more accurate and efficient.

WaferSegClassNet -- A Light-weight Network for Classification and Segmentation of Semiconductor Wafer Defects

As the integration density and design intricacy of semiconductor wafers increase, the magnitude and complexity of defects in them are also on the rise. Since the manual inspection of wafer defects is costly, an automated artificial intelligence (AI) based computer-vision approach is highly desired. The previous works on defect analysis have several limitations, such as low accuracy and the need for separate models for classification and segmentation. For analyzing mixed-type defects, some previous works require separately training one model for each defect type, which is non-scalable. In this paper, we present WaferSegClassNet (WSCN), a novel network based on encoder-decoder architecture. WSCN performs simultaneous classification and segmentation of both single and mixed-type wafer defects. WSCN uses a "shared encoder" for classification, and segmentation, which allows training WSCN end-to-end. We use N-pair contrastive loss to first pretrain the encoder and then use BCE-Dice loss for segmentation, and categorical cross-entropy loss for classification. Use of N-pair contrastive loss helps in better embedding representation in the latent dimension of wafer maps. WSCN has a model size of only 0.51MB and performs only 0.2M FLOPS. Thus, it is much lighter than other state-of-the-art models. Also, it requires only 150 epochs for convergence, compared to 4,000 epochs needed by a previous work. We evaluate our model on the MixedWM38 dataset, which has 38,015 images. WSCN achieves an average classification accuracy of 98.2% and a dice coefficient of 0.9999. We are the first to show segmentation results on the MixedWM38 dataset. The source code can be obtained from https://github.com/ckmvigil/WaferSegClassNet.

On Convergence of Gradient Descent Ascent: A Tight Local Analysis

Gradient Descent Ascent (GDA) methods are the mainstream algorithms for minimax optimization in generative adversarial networks (GANs). Convergence properties of GDA have drawn significant interest in the recent literature. Specifically, for $\min_{\mathbf{x}} \max_{\mathbf{y}} f(\mathbf{x};\mathbf{y})$ where $f$ is strongly-concave in $\mathbf{y}$ and possibly nonconvex in $\mathbf{x}$, (Lin et al., 2020) proved the convergence of GDA with a stepsize ratio $\eta_{\mathbf{y}}/\eta_{\mathbf{x}}=\Theta(\kappa^2)$ where $\eta_{\mathbf{x}}$ and $\eta_{\mathbf{y}}$ are the stepsizes for $\mathbf{x}$ and $\mathbf{y}$ and $\kappa$ is the condition number for $\mathbf{y}$. While this stepsize ratio suggests a slow training of the min player, practical GAN algorithms typically adopt similar stepsizes for both variables, indicating a wide gap between theoretical and empirical results. In this paper, we aim to bridge this gap by analyzing the \emph{local convergence} of general \emph{nonconvex-nonconcave} minimax problems. We demonstrate that a stepsize ratio of $\Theta(\kappa)$ is necessary and sufficient for local convergence of GDA to a Stackelberg Equilibrium, where $\kappa$ is the local condition number for $\mathbf{y}$. We prove a nearly tight convergence rate with a matching lower bound. We further extend the convergence guarantees to stochastic GDA and extra-gradient methods (EG). Finally, we conduct several numerical experiments to support our theoretical findings.

Tricking the Hashing Trick: A Tight Lower Bound on the Robustness of CountSketch to Adaptive Inputs

CountSketch and Feature Hashing (the "hashing trick") are popular randomized dimensionality reduction methods that support recovery of $\ell_2$-heavy hitters (keys $i$ where $v_i^2 > \epsilon \|\boldsymbol{v}\|_2^2$) and approximate inner products. When the inputs are {\em not adaptive} (do not depend on prior outputs), classic estimators applied to a sketch of size $O(\ell/\epsilon)$ are accurate for a number of queries that is exponential in $\ell$. When inputs are adaptive, however, an adversarial input can be constructed after $O(\ell)$ queries with the classic estimator and the best known robust estimator only supports $\tilde{O}(\ell^2)$ queries. In this work we show that this quadratic dependence is in a sense inherent: We design an attack that after $O(\ell^2)$ queries produces an adversarial input vector whose sketch is highly biased. Our attack uses "natural" non-adaptive inputs (only the final adversarial input is chosen adaptively) and universally applies with any correct estimator, including one that is unknown to the attacker. In that, we expose inherent vulnerability of this fundamental method.

M-Adapter: Modality Adaptation for End-to-End Speech-to-Text Translation

End-to-end speech-to-text translation models are often initialized with pre-trained speech encoder and pre-trained text decoder. This leads to a significant training gap between pre-training and fine-tuning, largely due to the modality differences between speech outputs from the encoder and text inputs to the decoder. In this work, we aim to bridge the modality gap between speech and text to improve translation quality. We propose M-Adapter, a novel Transformer-based module, to adapt speech representations to text. While shrinking the speech sequence, M-Adapter produces features desired for speech-to-text translation via modelling global and local dependencies of a speech sequence. Our experimental results show that our model outperforms a strong baseline by up to 1 BLEU score on the Must-C En$\rightarrow$DE dataset.\footnote{Our code is available at https://github.com/mingzi151/w2v2-st.}

Interpretable by Design: Learning Predictors by Composing Interpretable Queries

There is a growing concern about typically opaque decision-making with high-performance machine learning algorithms. Providing an explanation of the reasoning process in domain-specific terms can be crucial for adoption in risk-sensitive domains such as healthcare. We argue that machine learning algorithms should be interpretable by design and that the language in which these interpretations are expressed should be domain- and task-dependent. Consequently, we base our model's prediction on a family of user-defined and task-specific binary functions of the data, each having a clear interpretation to the end-user. We then minimize the expected number of queries needed for accurate prediction on any given input. As the solution is generally intractable, following prior work, we choose the queries sequentially based on information gain. However, in contrast to previous work, we need not assume the queries are conditionally independent. Instead, we leverage a stochastic generative model (VAE) and an MCMC algorithm (Unadjusted Langevin) to select the most informative query about the input based on previous query-answers. This enables the online determination of a query chain of whatever depth is required to resolve prediction ambiguities. Finally, experiments on vision and NLP tasks demonstrate the efficacy of our approach and its superiority over post-hoc explanations.

Wireless Channel Prediction in Partially Observed Environments

Site-specific radio frequency (RF) propagation prediction increasingly relies on models built from visual data such as cameras and LIDAR sensors. When operating in dynamic settings, the environment may only be partially observed. This paper introduces a method to extract statistical channel models, given partial observations of the surrounding environment. We propose a simple heuristic algorithm that performs ray tracing on the partial environment and then uses machine-learning trained predictors to estimate the channel and its uncertainty from features extracted from the partial ray tracing results. It is shown that the proposed method can interpolate between fully statistical models when no partial information is available and fully deterministic models when the environment is completely observed. The method can also capture the degree of uncertainty of the propagation predictions depending on the amount of region that has been explored. The methodology is demonstrated in a robotic navigation application simulated on a set of indoor maps with detailed models constructed using state-of-the-art navigation, simultaneous localization and mapping (SLAM), and computer vision methods.

Graph Learning based Generative Design for Resilience of Interdependent Network Systems

Interconnected complex systems usually undergo disruptions due to internal uncertainties and external negative impacts such as those caused by harsh operating environments or regional natural disaster events. To maintain the operation of interconnected network systems under both internal and external challenges, design for resilience research has been conducted from both enhancing the reliability of the system through better designs and improving the failure recovery capabilities. As for enhancing the designs, challenges have arisen for designing a robust system due to the increasing scale of modern systems and the complicated underlying physical constraints. To tackle these challenges and design a resilient system efficiently, this study presents a generative design method that utilizes graph learning algorithms. The generative design framework contains a performance estimator and a candidate design generator. The generator can intelligently mine good properties from existing systems and output new designs that meet predefined performance criteria. While the estimator can efficiently predict the performance of the generated design for a fast iterative learning process. Case studies results based on power systems from the IEEE dataset have illustrated the applicability of the proposed method for designing resilient interconnected systems.

An AlphaZero-Inspired Approach to Solving Search Problems

AlphaZero and its extension MuZero are computer programs that use machine-learning techniques to play at a superhuman level in chess, go, and a few other games. They achieved this level of play solely with reinforcement learning from self-play, without any domain knowledge except the game rules. It is a natural idea to adapt the methods and techniques used in AlphaZero for solving search problems such as the Boolean satisfiability problem (in its search version). Given a search problem, how to represent it for an AlphaZero-inspired solver? What are the "rules of solving" for this search problem? We describe possible representations in terms of easy-instance solvers and self-reductions, and we give examples of such representations for the satisfiability problem. We also describe a version of Monte Carlo tree search adapted for search problems.

Accelerating System-Level Debug Using Rule Learning and Subgroup Discovery Techniques

We propose a root-causing procedure for accelerating system-level debug using rule-based techniques. We describe the procedure and how it provides high quality debug hints for reducing the debug effort. This includes the heuristics for engineering features from logs of many tests, and the data analytics techniques for generating powerful debug hints. As a case study, we used these techniques for root-causing failures of the Power Management (PM) design feature Package-C8 and showed their effectiveness. Furthermore, we propose an approach for mining the root-causing experience and results for reuse, to accelerate future debug activities and reduce dependency on validation experts. We believe that these techniques are beneficial also for other validation activities at different levels of abstraction, for complex hardware, software and firmware systems, both pre-silicon and post-silicon.

SKIPP'D: a SKy Images and Photovoltaic Power Generation Dataset for Short-term Solar Forecasting

Large-scale integration of photovoltaics (PV) into electricity grids is challenged by the intermittent nature of solar power. Sky-image-based solar forecasting using deep learning has been recognized as a promising approach to predicting the short-term fluctuations. However, there are few publicly available standardized benchmark datasets for image-based solar forecasting, which limits the comparison of different forecasting models and the exploration of forecasting methods. To fill these gaps, we introduce SKIPP'D -- a SKy Images and Photovoltaic Power Generation Dataset. The dataset contains three years (2017-2019) of quality-controlled down-sampled sky images and PV power generation data that is ready-to-use for short-term solar forecasting using deep learning. In addition, to support the flexibility in research, we provide the high resolution, high frequency sky images and PV power generation data as well as the concurrent sky video footage. We also include a code base containing data processing scripts and baseline model implementations for researchers to reproduce our previous work and accelerate their research in solar forecasting.

A Structured Sparse Neural Network and Its Matrix Calculations Algorithm

Gradient descent optimizations and backpropagation are the most common methods for training neural networks, but they are computationally expensive for real time applications, need high memory resources, and are difficult to converge for many networks and large datasets. [Pseudo]inverse models for training neural network have emerged as powerful tools to overcome these issues. In order to effectively implement these methods, structured pruning maybe be applied to produce sparse neural networks. Although sparse neural networks are efficient in memory usage, most of their algorithms use the same fully loaded matrix calculation methods which are not efficient for sparse matrices. Tridiagonal matrices are one of the frequently used candidates for structuring neural networks, but they are not flexible enough to handle underfitting and overfitting problems as well as generalization properties. In this paper, we introduce a nonsymmetric, tridiagonal matrix with offdiagonal sparse entries and offset sub and super-diagonals as well algorithms for its [pseudo]inverse and determinant calculations. Traditional algorithms for matrix calculations, specifically inversion and determinant, of these forms are not efficient specially for large matrices, e.g. larger datasets or deeper networks. A decomposition for lower triangular matrices is developed and the original matrix is factorized into a set of matrices where their inverse matrices are calculated. For the cases where the matrix inverse does not exist, a least square type pseudoinverse is provided. The present method is a direct routine, i.e., executes in a predictable number of operations which is tested for randomly generated matrices with varying size. The results show significant improvement in computational costs specially when the size of matrix increases.

Reinforcement Learning Approaches for the Orienteering Problem with Stochastic and Dynamic Release Dates

In this paper, we study a sequential decision making problem faced by e-commerce carriers related to when to send out a vehicle from the central depot to serve customer requests, and in which order to provide the service, under the assumption that the time at which parcels arrive at the depot is stochastic and dynamic. The objective is to maximize the number of parcels that can be delivered during the service hours. We propose two reinforcement learning approaches for solving this problem, one based on a policy function approximation (PFA) and the second on a value function approximation (VFA). Both methods are combined with a look-ahead strategy, in which future release dates are sampled in a Monte-Carlo fashion and a tailored batch approach is used to approximate the value of future states. Our PFA and VFA make a good use of branch-and-cut-based exact methods to improve the quality of decisions. We also establish sufficient conditions for partial characterization of optimal policy and integrate them into PFA/VFA. In an empirical study based on 720 benchmark instances, we conduct a competitive analysis using upper bounds with perfect information and we show that PFA and VFA greatly outperform two alternative myopic approaches. Overall, PFA provides best solutions, while VFA (which benefits from a two-stage stochastic optimization model) achieves a better tradeoff between solution quality and computing time.

Tree ensemble kernels for Bayesian optimization with known constraints over mixed-feature spaces

Tree ensembles can be well-suited for black-box optimization tasks such as algorithm tuning and neural architecture search, as they achieve good predictive performance with little to no manual tuning, naturally handle discrete feature spaces, and are relatively insensitive to outliers in the training data. Two well-known challenges in using tree ensembles for black-box optimization are (i) effectively quantifying model uncertainty for exploration and (ii) optimizing over the piece-wise constant acquisition function. To address both points simultaneously, we propose using the kernel interpretation of tree ensembles as a Gaussian Process prior to obtain model variance estimates, and we develop a compatible optimization formulation for the acquisition function. The latter further allows us to seamlessly integrate known constraints to improve sampling efficiency by considering domain-knowledge in engineering settings and modeling search space symmetries, e.g., hierarchical relationships in neural architecture search. Our framework performs as well as state-of-the-art methods for unconstrained black-box optimization over continuous/discrete features and outperforms competing methods for problems combining mixed-variable feature spaces and known input constraints.

FL-Defender: Combating Targeted Attacks in Federated Learning

Federated learning (FL) enables learning a global machine learning model from local data distributed among a set of participating workers. This makes it possible i) to train more accurate models due to learning from rich joint training data, and ii) to improve privacy by not sharing the workers' local private data with others. However, the distributed nature of FL makes it vulnerable to targeted poisoning attacks that negatively impact the integrity of the learned model while, unfortunately, being difficult to detect. Existing defenses against those attacks are limited by assumptions on the workers' data distribution, may degrade the global model performance on the main task and/or are ill-suited to high-dimensional models. In this paper, we analyze targeted attacks against FL and find that the neurons in the last layer of a deep learning (DL) model that are related to the attacks exhibit a different behavior from the unrelated neurons, making the last-layer gradients valuable features for attack detection. Accordingly, we propose \textit{FL-Defender} as a method to combat FL targeted attacks. It consists of i) engineering more robust discriminative features by calculating the worker-wise angle similarity for the workers' last-layer gradients, ii) compressing the resulting similarity vectors using PCA to reduce redundant information, and iii) re-weighting the workers' updates based on their deviation from the centroid of the compressed similarity vectors. Experiments on three data sets with different DL model sizes and data distributions show the effectiveness of our method at defending against label-flipping and backdoor attacks. Compared to several state-of-the-art defenses, FL-Defender achieves the lowest attack success rates, maintains the performance of the global model on the main task and causes minimal computational overhead on the server.

The Linguistic Blind Spot of Value-Aligned Agency, Natural and Artificial

The value-alignment problem for artificial intelligence (AI) asks how we can ensure that the 'values' (i.e., objective functions) of artificial systems are aligned with the values of humanity. In this paper, I argue that linguistic communication (natural language) is a necessary condition for robust value alignment. I discuss the consequences that the truth of this claim would have for research programmes that attempt to ensure value alignment for AI systems; or, more loftily, designing robustly beneficial or ethical artificial agents.

Combinatory Adjoints and Differentiation

We develop a compositional approach for automatic and symbolic differentiation based on categorical constructions in functional analysis where derivatives are linear functions on abstract vectors rather than being limited to scalars, vectors, matrices or tensors represented as multi-dimensional arrays. We show that both symbolic and automatic differentiation can be performed using a differential calculus for generating linear functions representing Fr\'echet derivatives based on rules for primitive, constant, linear and bilinear functions as well as their sequential and parallel composition. Linear functions are represented in a combinatory domain-specific language. Finally, we provide a calculus for symbolically computing the adjoint of a derivative without using matrices, which are too inefficient to use on high-dimensional spaces. The resulting symbolic representation of a derivative retains the data-parallel operations from the input program. The combination of combinatory differentiation and computing formal adjoints turns out to be behaviorally equivalent to reverse-mode automatic differentiation. In particular, it provides opportunities for optimizations where matrices are too inefficient to represent linear functions.

Less Is More: A Comparison of Active Learning Strategies for 3D Medical Image Segmentation

Since labeling medical image data is a costly and labor-intensive process, active learning has gained much popularity in the medical image segmentation domain in recent years. A variety of active learning strategies have been proposed in the literature, but their effectiveness is highly dependent on the dataset and training scenario. To facilitate the comparison of existing strategies and provide a baseline for evaluating novel strategies, we evaluate the performance of several well-known active learning strategies on three datasets from the Medical Segmentation Decathlon. Additionally, we consider a strided sampling strategy specifically tailored to 3D image data. We demonstrate that both random and strided sampling act as strong baselines and discuss the advantages and disadvantages of the studied methods. To allow other researchers to compare their work to our results, we provide an open-source framework for benchmarking active learning strategies on a variety of medical segmentation datasets.

GOF-TTE: Generative Online Federated Learning Framework for Travel Time Estimation

Estimating the travel time of a path is an essential topic for intelligent transportation systems. It serves as the foundation for real-world applications, such as traffic monitoring, route planning, and taxi dispatching. However, building a model for such a data-driven task requires a large amount of users' travel information, which directly relates to their privacy and thus is less likely to be shared. The non-Independent and Identically Distributed (non-IID) trajectory data across data owners also make a predictive model extremely challenging to be personalized if we directly apply federated learning. Finally, previous work on travel time estimation does not consider the real-time traffic state of roads, which we argue can significantly influence the prediction. To address the above challenges, we introduce GOF-TTE for the mobile user group, Generative Online Federated Learning Framework for Travel Time Estimation, which I) utilizes the federated learning approach, allowing private data to be kept on client devices while training, and designs the global model as an online generative model shared by all clients to infer the real-time road traffic state. II) apart from sharing a base model at the server, adapts a fine-tuned personalized model for every client to study their personal driving habits, making up for the residual error made by localized global model prediction. % III) designs the global model as an online generative model shared by all clients to infer the real-time road traffic state. We also employ a simple privacy attack to our framework and implement the differential privacy mechanism to further guarantee privacy safety. Finally, we conduct experiments on two real-world public taxi datasets of DiDi Chengdu and Xi'an. The experimental results demonstrate the effectiveness of our proposed framework.

A Multi-Task BERT Model for Schema-Guided Dialogue State Tracking

Task-oriented dialogue systems often employ a Dialogue State Tracker (DST) to successfully complete conversations. Recent state-of-the-art DST implementations rely on schemata of diverse services to improve model robustness and handle zero-shot generalization to new domains [1], however such methods [2, 3] typically require multiple large scale transformer models and long input sequences to perform well. We propose a single multi-task BERT-based model that jointly solves the three DST tasks of intent prediction, requested slot prediction and slot filling. Moreover, we propose an efficient and parsimonious encoding of the dialogue history and service schemata that is shown to further improve performance. Evaluation on the SGD dataset shows that our approach outperforms the baseline SGP-DST by a large margin and performs well compared to the state-of-the-art, while being significantly more computationally efficient. Extensive ablation studies are performed to examine the contributing factors to the success of our model.

Firenze: Model Evaluation Using Weak Signals

Data labels in the security field are frequently noisy, limited, or biased towards a subset of the population. As a result, commonplace evaluation methods such as accuracy, precision and recall metrics, or analysis of performance curves computed from labeled datasets do not provide sufficient confidence in the real-world performance of a machine learning (ML) model. This has slowed the adoption of machine learning in the field. In the industry today, we rely on domain expertise and lengthy manual evaluation to build this confidence before shipping a new model for security applications. In this paper, we introduce Firenze, a novel framework for comparative evaluation of ML models' performance using domain expertise, encoded into scalable functions called markers. We show that markers computed and combined over select subsets of samples called regions of interest can provide a robust estimate of their real-world performances. Critically, we use statistical hypothesis testing to ensure that observed differences-and therefore conclusions emerging from our framework-are more prominent than that observable from the noise alone. Using simulations and two real-world datasets for malware and domain-name-service reputation detection, we illustrate our approach's effectiveness, limitations, and insights. Taken together, we propose Firenze as a resource for fast, interpretable, and collaborative model development and evaluation by mixed teams of researchers, domain experts, and business owners.

Lane-GNN: Integrating GNN for Predicting Drivers' Lane Change Intention

Nowadays, intelligent highway traffic network is playing an important role in modern transportation infrastructures. A variable speed limit (VSL) system can be facilitated in the highway traffic network to provide useful and dynamic speed limit information for drivers to travel with enhanced safety. Such system is usually designed with a steady advisory speed in mind so that traffic can move smoothly when drivers follow the speed, rather than speeding up whenever there is a gap and slowing down at congestion. However, little attention has been given to the research of vehicles' behaviours when drivers left the road network governed by a VSL system, which may largely involve unexpected acceleration, deceleration and frequent lane changes, resulting in chaos for the subsequent highway road users. In this paper, we focus on the detection of traffic flow anomaly due to drivers' lane change intention on the highway traffic networks after a VSL system. More specifically, we apply graph modelling on the traffic flow data generated by a popular mobility simulator, SUMO, at road segment levels. We then evaluate the performance of lane changing detection using the proposed Lane-GNN scheme, an attention temporal graph convolutional neural network, and compare its performance with a temporal convolutional neural network (TCNN) as our baseline. Our experimental results show that the proposed Lane-GNN can detect drivers' lane change intention within 90 seconds with an accuracy of 99.42% under certain assumptions. Finally, some interpretation methods are applied to the trained models with a view to further illustrate our findings.

PGMG: A Pharmacophore-Guided Deep Learning Approach for Bioactive Molecular Generation

The rational design of novel molecules with desired bioactivity is a critical but challenging task in drug discovery, especially when treating a novel target family or understudied targets. Here, we propose PGMG, a pharmacophore-guided deep learning approach for bioactivate molecule generation. Through the guidance of pharmacophore, PGMG provides a flexible strategy to generate bioactive molecules with structural diversity in various scenarios using a trained variational autoencoder. We show that PGMG can generate molecules matching given pharmacophore models while maintaining a high level of validity, uniqueness, and novelty. In the case studies, we demonstrate the application of PGMG to generate bioactive molecules in ligand-based and structure-based drug de novo design, as well as in lead optimization scenarios. Overall, the flexibility and effectiveness of PGMG make it a useful tool for accelerating the drug discovery process.

Geometric Learning of Hidden Markov Models via a Method of Moments Algorithm

We present a novel algorithm for learning the parameters of hidden Markov models (HMMs) in a geometric setting where the observations take values in Riemannian manifolds. In particular, we elevate a recent second-order method of moments algorithm that incorporates non-consecutive correlations to a more general setting where observations take place in a Riemannian symmetric space of non-positive curvature and the observation likelihoods are Riemannian Gaussians. The resulting algorithm decouples into a Riemannian Gaussian mixture model estimation algorithm followed by a sequence of convex optimization procedures. We demonstrate through examples that the learner can result in significantly improved speed and numerical accuracy compared to existing learners.

Biologically-informed deep learning models for cancer: fundamental trends for encoding and interpreting oncology data

In this paper we provide a structured literature analysis focused on Deep Learning (DL) models used to support inference in cancer biology with a particular emphasis on multi-omics analysis. The work focuses on how existing models address the need for better dialogue with prior knowledge, biological plausibility and interpretability, fundamental properties in the biomedical domain. We discuss the recent evolutionary arch of DL models in the direction of integrating prior biological relational and network knowledge to support better generalisation (e.g. pathways or Protein-Protein-Interaction networks) and interpretability. This represents a fundamental functional shift towards models which can integrate mechanistic and statistical inference aspects. We discuss representational methodologies for the integration of domain prior knowledge in such models. The paper also provides a critical outlook into contemporary methods for explainability and interpretabiltiy. This analysis points in the direction of a convergence between encoding prior knowledge and improved interpretability.

Eliciting and Learning with Soft Labels from Every Annotator

The labels used to train machine learning (ML) models are of paramount importance. Typically for ML classification tasks, datasets contain hard labels, yet learning using soft labels has been shown to yield benefits for model generalization, robustness, and calibration. Earlier work found success in forming soft labels from multiple annotators' hard labels; however, this approach may not converge to the best labels and necessitates many annotators, which can be expensive and inefficient. We focus on efficiently eliciting soft labels from individual annotators. We collect and release a dataset of soft labels for CIFAR-10 via a crowdsourcing study ($N=242$). We demonstrate that learning with our labels achieves comparable model performance to prior approaches while requiring far fewer annotators. Our elicitation methodology therefore shows promise towards enabling practitioners to enjoy the benefits of improved model performance and reliability with fewer annotators, and serves as a guide for future dataset curators on the benefits of leveraging richer information, such as categorical uncertainty, from individual annotators.

On the modern deep learning approaches for precipitation downscaling

Deep Learning (DL) based downscaling has become a popular tool in earth sciences recently. Increasingly, different DL approaches are being adopted to downscale coarser precipitation data and generate more accurate and reliable estimates at local (~few km or even smaller) scales. Despite several studies adopting dynamical or statistical downscaling of precipitation, the accuracy is limited by the availability of ground truth. A key challenge to gauge the accuracy of such methods is to compare the downscaled data to point-scale observations which are often unavailable at such small scales. In this work, we carry out the DL-based downscaling to estimate the local precipitation data from the India Meteorological Department (IMD), which was created by approximating the value from station location to a grid point. To test the efficacy of different DL approaches, we apply four different methods of downscaling and evaluate their performance. The considered approaches are (i) Deep Statistical Downscaling (DeepSD), augmented Convolutional Long Short Term Memory (ConvLSTM), fully convolutional network (U-NET), and Super-Resolution Generative Adversarial Network (SR-GAN). A custom VGG network, used in the SR-GAN, is developed in this work using precipitation data. The results indicate that SR-GAN is the best method for precipitation data downscaling. The downscaled data is validated with precipitation values at IMD station. This DL method offers a promising alternative to statistical downscaling.

An AIoT-enabled Autonomous Dementia Monitoring System

An autonomous Artificial Internet of Things (AIoT) system for elderly dementia patients monitoring in a smart home is presented. The system mainly implements two functions based on the activity inference of the sensor data, which are real time abnormal activity monitoring and trend prediction of disease related activities. Specifically, CASAS dataset is employed to train a Random Forest (RF) model for activity inference. Then, another RF model trained by the output data of activity inference is used for abnormal activity monitoring. Particularly, RF is chosen for these tasks because of its balanced trade offs between accuracy, time efficiency, flexibility, and interpretability. Moreover, Long Short Term Memory (LSTM) is utilised to forecast the disease related activity trend of a patient. Consequently, the accuracy of two RF classifiers designed for activity inference and abnormal activity detection is greater than 99 percent and 94 percent, respectively. Furthermore, using the duration of sleep as an example, the LSTM model achieves accurate and evident future trends prediction.

Object Representations as Fixed Points: Training Iterative Refinement Algorithms with Implicit Differentiation

Iterative refinement -- start with a random guess, then iteratively improve the guess -- is a useful paradigm for representation learning because it offers a way to break symmetries among equally plausible explanations for the data. This property enables the application of such methods to infer representations of sets of entities, such as objects in physical scenes, structurally resembling clustering algorithms in latent space. However, most prior works differentiate through the unrolled refinement process, which can make optimization challenging. We observe that such methods can be made differentiable by means of the implicit function theorem, and develop an implicit differentiation approach that improves the stability and tractability of training by decoupling the forward and backward passes. This connection enables us to apply advances in optimizing implicit layers to not only improve the optimization of the slot attention module in SLATE, a state-of-the-art method for learning entity representations, but do so with constant space and time complexity in backpropagation and only one additional line of code.

ANEC: An Amharic Named Entity Corpus and Transformer Based Recognizer

Named Entity Recognition is an information extraction task that serves as a preprocessing step for other natural language processing tasks, such as machine translation, information retrieval, and question answering. Named entity recognition enables the identification of proper names as well as temporal and numeric expressions in an open domain text. For Semitic languages such as Arabic, Amharic, and Hebrew, the named entity recognition task is more challenging due to the heavily inflected structure of these languages. In this paper, we present an Amharic named entity recognition system based on bidirectional long short-term memory with a conditional random fields layer. We annotate a new Amharic named entity recognition dataset (8,070 sentences, which has 182,691 tokens) and apply Synthetic Minority Over-sampling Technique to our dataset to mitigate the imbalanced classification problem. Our named entity recognition system achieves an F_1 score of 93%, which is the new state-of-the-art result for Amharic named entity recognition.

Computer-assisted Pronunciation Training -- Speech synthesis is almost all you need

The research community has long studied computer-assisted pronunciation training (CAPT) methods in non-native speech. Researchers focused on studying various model architectures, such as Bayesian networks and deep learning methods, as well as on the analysis of different representations of the speech signal. Despite significant progress in recent years, existing CAPT methods are not able to detect pronunciation errors with high accuracy (only 60\% precision at 40\%-80\% recall). One of the key problems is the low availability of mispronounced speech that is needed for the reliable training of pronunciation error detection models. If we had a generative model that could mimic non-native speech and produce any amount of training data, then the task of detecting pronunciation errors would be much easier. We present three innovative techniques based on phoneme-to-phoneme (P2P), text-to-speech (T2S), and speech-to-speech (S2S) conversion to generate correctly pronounced and mispronounced synthetic speech. We show that these techniques not only improve the accuracy of three machine learning models for detecting pronunciation errors but also help establish a new state-of-the-art in the field. Earlier studies have used simple speech generation techniques such as P2P conversion, but only as an additional mechanism to improve the accuracy of pronunciation error detection. We, on the other hand, consider speech generation to be the first-class method of detecting pronunciation errors. The effectiveness of these techniques is assessed in the tasks of detecting pronunciation and lexical stress errors. Non-native English speech corpora of German, Italian, and Polish speakers are used in the evaluations. The best proposed S2S technique improves the accuracy of detecting pronunciation errors in AUC metric by 41\% from 0.528 to 0.749 compared to the state-of-the-art approach.

Unsupervised Symbolic Music Segmentation using Ensemble Temporal Prediction Errors

Symbolic music segmentation is the process of dividing symbolic melodies into smaller meaningful groups, such as melodic phrases. We proposed an unsupervised method for segmenting symbolic music. The proposed model is based on an ensemble of temporal prediction error models. During training, each model predicts the next token to identify musical phrase changes. While at test time, we perform a peak detection algorithm to select segment candidates. Finally, we aggregate the predictions of each of the models participating in the ensemble to predict the final segmentation. Results suggest the proposed method reaches state-of-the-art performance on the Essen Folksong dataset under the unsupervised setting when considering F-Score and R-value. We additionally provide an ablation study to better assess the contribution of each of the model components to the final results. As expected, the proposed method is inferior to the supervised setting, which leaves room for improvement in future research considering closing the gap between unsupervised and supervised methods.

Abstraction and Refinement: Towards Scalable and Exact Verification of Neural Networks

As a new programming paradigm, deep neural networks (DNNs) have been increasingly deployed in practice, but the lack of robustness hinders their applications in safety-critical domains. While there are techniques for verifying DNNs with formal guarantees, they are limited in scalability and accuracy. In this paper, we present a novel abstraction-refinement approach for scalable and exact DNN verification. Specifically, we propose a novel abstraction to break down the size of DNNs by over-approximation. The result of verifying the abstract DNN is always conclusive if no spurious counterexample is reported. To eliminate spurious counterexamples introduced by abstraction, we propose a novel counterexample-guided refinement that refines the abstract DNN to exclude a given spurious counterexample while still over-approximating the original one. Our approach is orthogonal to and can be integrated with many existing verification techniques. For demonstration, we implement our approach using two promising and exact tools Marabou and Planet as the underlying verification engines, and evaluate on widely-used benchmarks ACAS Xu, MNIST and CIFAR-10. The results show that our approach can boost their performance by solving more problems and reducing up to 86.3% and 78.0% verification time, respectively. Compared to the most relevant abstraction-refinement approach, our approach is 11.6-26.6 times faster.

Unsupervised Recurrent Federated Learning for Edge Popularity Prediction in Privacy-Preserving Mobile Edge Computing Networks

Nowadays wireless communication is rapidly reshaping entire industry sectors. In particular, mobile edge computing (MEC) as an enabling technology for industrial Internet of things (IIoT) brings powerful computing/storage infrastructure closer to the mobile terminals and, thereby, significant lowers the response latency. To reap the benefit of proactive caching at the network edge, precise knowledge on the popularity pattern among the end devices is essential. However, the complex and dynamic nature of the content popularity over space and time as well as the data-privacy requirements in many IIoT scenarios pose tough challenges to its acquisition. In this article, we propose an unsupervised and privacy-preserving popularity prediction framework for MEC-enabled IIoT. The concepts of local and global popularities are introduced and the time-varying popularity of each user is modelled as a model-free Markov chain. On this basis, a novel unsupervised recurrent federated learning (URFL) algorithm is proposed to predict the distributed popularity while achieve privacy preservation and unsupervised training. Simulations indicate that the proposed framework can enhance the prediction accuracy in terms of a reduced root-mean-squared error by up to $60.5\%-68.7\%$. Additionally, manual labeling and violation of users' data privacy are both avoided.

Informed Learning by Wide Neural Networks: Convergence, Generalization and Sampling Complexity

By integrating domain knowledge with labeled samples, informed machine learning has been emerging to improve the learning performance for a wide range of applications. Nonetheless, rigorous understanding of the role of injected domain knowledge has been under-explored. In this paper, we consider an informed deep neural network (DNN) with over-parameterization and domain knowledge integrated into its training objective function, and study how and why domain knowledge benefits the performance. Concretely, we quantitatively demonstrate the two benefits of domain knowledge in informed learning - regularizing the label-based supervision and supplementing the labeled samples - and reveal the trade-off between label and knowledge imperfectness in the bound of the population risk. Based on the theoretical analysis, we propose a generalized informed training objective to better exploit the benefits of knowledge and balance the label and knowledge imperfectness, which is validated by the population risk bound. Our analysis on sampling complexity sheds lights on how to choose the hyper-parameters for informed learning, and further justifies the advantages of knowledge informed learning.

Scheduling Planting Time Through Developing an Optimization Model and Analysis of Time Series Growing Degree Units

Producing higher-quality crops within shortened breeding cycles ensures global food availability and security, but this improvement intensifies logistical and productivity challenges for seed industries in the year-round breeding process due to the storage limitations. In the 2021 Syngenta crop challenge in analytics, Syngenta raised the problem to design an optimization model for the planting time scheduling in the 2020 year-round breeding process so that there is a consistent harvest quantity each week. They released a dataset that contained 2569 seed populations with their planting windows, required growing degree units for harvesting, and their harvest quantities at two sites. To address this challenge, we developed a new framework that consists of a weather time series model and an optimization model to schedule the planting time. A deep recurrent neural network was designed to predict the weather into the future, and a Gaussian process model on top of the time-series model was developed to model the uncertainty of forecasted weather. The proposed optimization models also scheduled the seed population's planting time at the fewest number of weeks with a more consistent weekly harvest quantity. Using the proposed optimization models can decrease the required capacity by 69% at site 0 and up to 51% at site 1 compared to the original planting time.

PhilaeX: Explaining the Failure and Success of AI Models in Malware Detection

The explanation to an AI model's prediction used to support decision making in cyber security, is of critical importance. It is especially so when the model's incorrect prediction can lead to severe damages or even losses to lives and critical assets. However, most existing AI models lack the ability to provide explanations on their prediction results, despite their strong performance in most scenarios. In this work, we propose a novel explainable AI method, called PhilaeX, that provides the heuristic means to identify the optimized subset of features to form the complete explanations of AI models' predictions. It identifies the features that lead to the model's borderline prediction, and those with positive individual contributions are extracted. The feature attributions are then quantified through the optimization of a Ridge regression model. We verify the explanation fidelity through two experiments. First, we assess our method's capability in correctly identifying the activated features in the adversarial samples of Android malwares, through the features attribution values from PhilaeX. Second, the deduction and augmentation tests, are used to assess the fidelity of the explanations. The results show that PhilaeX is able to explain different types of classifiers correctly, with higher fidelity explanations, compared to the state-of-the-arts methods such as LIME and SHAP.

Deep Learning for Systemic Risk Measures

The aim of this paper is to study a new methodological framework for systemic risk measures by applying deep learning method as a tool to compute the optimal strategy of capital allocations. Under this new framework, systemic risk measures can be interpreted as the minimal amount of cash that secures the aggregated system by allocating capital to the single institutions before aggregating the individual risks. This problem has no explicit solution except in very limited situations. Deep learning is increasingly receiving attention in financial modelings and risk management and we propose our deep learning based algorithms to solve both the primal and dual problems of the risk measures, and thus to learn the fair risk allocations. In particular, our method for the dual problem involves the training philosophy inspired by the well-known Generative Adversarial Networks (GAN) approach and a newly designed direct estimation of Radon-Nikodym derivative. We close the paper with substantial numerical studies of the subject and provide interpretations of the risk allocations associated to the systemic risk measures. In the particular case of exponential preferences, numerical experiments demonstrate excellent performance of the proposed algorithm, when compared with the optimal explicit solution as a benchmark.

SketchCleanNet -- A deep learning approach to the enhancement and correction of query sketches for a 3D CAD model retrieval system

Search and retrieval remains a major research topic in several domains, including computer graphics, computer vision, engineering design, etc. A search engine requires primarily an input search query and a database of items to search from. In engineering, which is the primary context of this paper, the database consists of 3D CAD models, such as washers, pistons, connecting rods, etc. A query from a user is typically in the form of a sketch, which attempts to capture the details of a 3D model. However, sketches have certain typical defects such as gaps, over-drawn portions (multi-strokes), etc. Since the retrieved results are only as good as the input query, sketches need cleaning-up and enhancement for better retrieval results. In this paper, a deep learning approach is proposed to improve or clean the query sketches. Initially, sketches from various categories are analysed in order to understand the many possible defects that may occur. A dataset of cleaned-up or enhanced query sketches is then created based on an understanding of these defects. Consequently, an end-to-end training of a deep neural network is carried out in order to provide a mapping between the defective and the clean sketches. This network takes the defective query sketch as the input and generates a clean or an enhanced query sketch. Qualitative and quantitative comparisons of the proposed approach with other state-of-the-art techniques show that the proposed approach is effective. The results of the search engine are reported using both the defective and enhanced query sketches, and it is shown that using the enhanced query sketches from the developed approach yields improved search results.

q-Learning in Continuous Time

We study the continuous-time counterpart of Q-learning for reinforcement learning (RL) under the entropy-regularized, exploratory diffusion process formulation introduced by Wang et al. (2020) As the conventional (big) Q-function collapses in continuous time, we consider its first-order approximation and coin the term "(little) q-function". This function is related to the instantaneous advantage rate function as well as the Hamiltonian. We develop a "q-learning" theory around the q-function that is independent of time discretization. Given a stochastic policy, we jointly characterize the associated q-function and value function by martingale conditions of certain stochastic processes. We then apply the theory to devise different actor-critic algorithms for solving underlying RL problems, depending on whether or not the density function of the Gibbs measure generated from the q-function can be computed explicitly. One of our algorithms interprets the well-known Q-learning algorithm SARSA, and another recovers a policy gradient (PG) based continuous-time algorithm proposed in Jia and Zhou (2021). Finally, we conduct simulation experiments to compare the performance of our algorithms with those of PG-based algorithms in Jia and Zhou (2021) and time-discretized conventional Q-learning algorithms.

Parameter efficient dendritic-tree neurons outperform perceptrons

Biological neurons are more powerful than artificial perceptrons, in part due to complex dendritic input computations. Inspired to empower the perceptron with biologically inspired features, we explore the effect of adding and tuning input branching factors along with input dropout. This allows for parameter efficient non-linear input architectures to be discovered and benchmarked. Furthermore, we present a PyTorch module to replace multi-layer perceptron layers in existing architectures. Our initial experiments on MNIST classification demonstrate the accuracy and generalization improvement of dendritic neurons compared to existing perceptron architectures.

UserLibri: A Dataset for ASR Personalization Using Only Text

Personalization of speech models on mobile devices (on-device personalization) is an active area of research, but more often than not, mobile devices have more text-only data than paired audio-text data. We explore training a personalized language model on text-only data, used during inference to improve speech recognition performance for that user. We experiment on a user-clustered LibriSpeech corpus, supplemented with personalized text-only data for each user from Project Gutenberg. We release this User-Specific LibriSpeech (UserLibri) dataset to aid future personalization research. LibriSpeech audio-transcript pairs are grouped into 55 users from the test-clean dataset and 52 users from test-other. We are able to lower the average word error rate per user across both sets in streaming and nonstreaming models, including an improvement of 2.5 for the harder set of test-other users when streaming.

Multivariate Time Series Anomaly Detection with Few Positive Samples

Given the scarcity of anomalies in real-world applications, the majority of literature has been focusing on modeling normality. The learned representations enable anomaly detection as the normality model is trained to capture certain key underlying data regularities under normal circumstances. In practical settings, particularly industrial time series anomaly detection, we often encounter situations where a large amount of normal operation data is available along with a small number of anomaly events collected over time. This practical situation calls for methodologies to leverage these small number of anomaly events to create a better anomaly detector. In this paper, we introduce two methodologies to address the needs of this practical situation and compared them with recently developed state of the art techniques. Our proposed methods anchor on representative learning of normal operation with autoregressive (AR) model along with loss components to encourage representations that separate normal versus few positive examples. We applied the proposed methods to two industrial anomaly detection datasets and demonstrated effective performance in comparison with approaches from literature. Our study also points out additional challenges with adopting such methods in practical applications.

Uncertainty Quantification for Deep Unrolling-Based Computational Imaging

Deep unrolling is an emerging deep learning-based image reconstruction methodology that bridges the gap between model-based and purely deep learning-based image reconstruction methods. Although deep unrolling methods achieve state-of-the-art performance for imaging problems and allow the incorporation of the observation model into the reconstruction process, they do not provide any uncertainty information about the reconstructed image, which severely limits their use in practice, especially for safety-critical imaging applications. In this paper, we propose a learning-based image reconstruction framework that incorporates the observation model into the reconstruction task and that is capable of quantifying epistemic and aleatoric uncertainties, based on deep unrolling and Bayesian neural networks. We demonstrate the uncertainty characterization capability of the proposed framework on magnetic resonance imaging and computed tomography reconstruction problems. We investigate the characteristics of the epistemic and aleatoric uncertainty information provided by the proposed framework to motivate future research on utilizing uncertainty information to develop more accurate, robust, trustworthy, uncertainty-aware, learning-based image reconstruction and analysis methods for imaging problems. We show that the proposed framework can provide uncertainty information while achieving comparable reconstruction performance to state-of-the-art deep unrolling methods.

Efficient Adversarial Training With Data Pruning

Neural networks are susceptible to adversarial examples-small input perturbations that cause models to fail. Adversarial training is one of the solutions that stops adversarial examples; models are exposed to attacks during training and learn to be resilient to them. Yet, such a procedure is currently expensive-it takes a long time to produce and train models with adversarial samples, and, what is worse, it occasionally fails. In this paper we demonstrate data pruning-a method for increasing adversarial training efficiency through data sub-sampling.We empirically show that data pruning leads to improvements in convergence and reliability of adversarial training, albeit with different levels of utility degradation. For example, we observe that using random sub-sampling of CIFAR10 to drop 40% of data, we lose 8% adversarial accuracy against the strongest attackers, while by using only 20% of data we lose 14% adversarial accuracy and reduce runtime by a factor of 3. Interestingly, we discover that in some settings data pruning brings benefits from both worlds-it both improves adversarial accuracy and training time.

Few-shot incremental learning in the context of solar cell quality inspection

In industry, Deep Neural Networks have shown high defect detection rates surpassing other more traditional manual feature engineering based proposals. This has been achieved mainly through supervised training where a great amount of data is required in order to learn good classification models. However, such amount of data is sometimes hard to obtain in industrial scenarios, as few defective pieces are produced normally. In addition, certain kinds of defects are very rare and usually just appear from time to time, which makes the generation of a proper dataset for training a classification model even harder. Moreover, the lack of available data limits the adaptation of inspection models to new defect types that appear in production as it might require a model retraining in order to incorporate the detects and detect them. In this work, we have explored the technique of weight imprinting in the context of solar cell quality inspection where we have trained a network on three base defect classes, and then we have incorporated new defect classes using few samples. The results have shown that this technique allows the network to extend its knowledge with regard to defect classes with few samples, which can be interesting for industrial practitioners.

American == White in Multimodal Language-and-Image AI

Three state-of-the-art language-and-image AI models, CLIP, SLIP, and BLIP, are evaluated for evidence of a bias previously observed in social and experimental psychology: equating American identity with being White. Embedding association tests (EATs) using standardized images of self-identified Asian, Black, Latina/o, and White individuals from the Chicago Face Database (CFD) reveal that White individuals are more associated with collective in-group words than are Asian, Black, or Latina/o individuals. In assessments of three core aspects of American identity reported by social psychologists, single-category EATs reveal that images of White individuals are more associated with patriotism and with being born in America, but that, consistent with prior findings in psychology, White individuals are associated with being less likely to treat people of all races and backgrounds equally. Three downstream machine learning tasks demonstrate biases associating American with White. In a visual question answering task using BLIP, 97% of White individuals are identified as American, compared to only 3% of Asian individuals. When asked in what state the individual depicted lives in, the model responds China 53% of the time for Asian individuals, but always with an American state for White individuals. In an image captioning task, BLIP remarks upon the race of Asian individuals as much as 36% of the time, but never remarks upon race for White individuals. Finally, provided with an initialization image from the CFD and the text "an American person," a synthetic image generator (VQGAN) using the text-based guidance of CLIP lightens the skin tone of individuals of all races (by 35% for Black individuals, based on pixel brightness). The results indicate that biases equating American identity with being White are learned by language-and-image AI, and propagate to downstream applications of such models.

Transforming PageRank into an Infinite-Depth Graph Neural Network

Popular graph neural networks are shallow models, despite the success of very deep architectures in other application domains of deep learning. This reduces the modeling capacity and leaves models unable to capture long-range relationships. The primary reason for the shallow design results from over-smoothing, which leads node states to become more similar with increased depth. We build on the close connection between GNNs and PageRank, for which personalized PageRank introduces the consideration of a personalization vector. Adopting this idea, we propose the Personalized PageRank Graph Neural Network (PPRGNN), which extends the graph convolutional network to an infinite-depth model that has a chance to reset the neighbor aggregation back to the initial state in each iteration. We introduce a nicely interpretable tweak to the chance of resetting and prove the convergence of our approach to a unique solution without placing any constraints, even when taking infinitely many neighbor aggregations. As in personalized PageRank, our result does not suffer from over-smoothing. While doing so, time complexity remains linear while we keep memory complexity constant, independently of the depth of the network, making it scale well to large graphs. We empirically show the effectiveness of our approach for various node and graph classification tasks. PPRGNN outperforms comparable methods in almost all cases.

Infinite-Fidelity Coregionalization for Physical Simulation

Multi-fidelity modeling and learning are important in physical simulation-related applications. It can leverage both low-fidelity and high-fidelity examples for training so as to reduce the cost of data generation while still achieving good performance. While existing approaches only model finite, discrete fidelities, in practice, the fidelity choice is often continuous and infinite, which can correspond to a continuous mesh spacing or finite element length. In this paper, we propose Infinite Fidelity Coregionalization (IFC). Given the data, our method can extract and exploit rich information within continuous, infinite fidelities to bolster the prediction accuracy. Our model can interpolate and/or extrapolate the predictions to novel fidelities, which can be even higher than the fidelities of training data. Specifically, we introduce a low-dimensional latent output as a continuous function of the fidelity and input, and multiple it with a basis matrix to predict high-dimensional solution outputs. We model the latent output as a neural Ordinary Differential Equation (ODE) to capture the complex relationships within and integrate information throughout the continuous fidelities. We then use Gaussian processes or another ODE to estimate the fidelity-varying bases. For efficient inference, we reorganize the bases as a tensor, and use a tensor-Gaussian variational posterior to develop a scalable inference algorithm for massive outputs. We show the advantage of our method in several benchmark tasks in computational physics.

DRESS: Dynamic REal-time Sparse Subnets

The limited and dynamically varied resources on edge devices motivate us to deploy an optimized deep neural network that can adapt its sub-networks to fit in different resource constraints. However, existing works often build sub-networks through searching different network architectures in a hand-crafted sampling space, which not only can result in a subpar performance but also may cause on-device re-configuration overhead. In this paper, we propose a novel training algorithm, Dynamic REal-time Sparse Subnets (DRESS). DRESS samples multiple sub-networks from the same backbone network through row-based unstructured sparsity, and jointly trains these sub-networks in parallel with weighted loss. DRESS also exploits strategies including parameter reusing and row-based fine-grained sampling for efficient storage consumption and efficient on-device adaptation. Extensive experiments on public vision datasets show that DRESS yields significantly higher accuracy than state-of-the-art sub-networks.

Improving Low-Resource Speech Recognition with Pretrained Speech Models: Continued Pretraining vs. Semi-Supervised Training

Self-supervised Transformer based models, such as wav2vec 2.0 and HuBERT, have produced significant improvements over existing approaches to automatic speech recognition (ASR). This is evident in the performance of the wav2vec 2.0 based pretrained XLSR-53 model across many languages when fine-tuned with available labeled data. However, the performance from finetuning these models can be dependent on the amount of in-language or similar-to-in-language data included in the pretraining dataset. In this paper we investigate continued pretraining (CoPT) with unlabeled in-language audio data on the XLSR-53 pretrained model in several low-resource languages. CoPT is more computationally efficient than semi-supervised training (SST), the standard approach of utilizing unlabeled data in ASR, since it omits the need for pseudo-labeling of the unlabeled data. We show CoPT results in word error rates (WERs), equal to or slightly better than using SST. In addition, we show that using the CoPT model for pseudo-labeling, and using these labels in SST, results in further improvements in WER.

Efficient Adaptive Regret Minimization

In online convex optimization the player aims to minimize her regret against a fixed comparator over the entire repeated game. Algorithms that minimize standard regret may converge to a fixed decision, which is undesireable in changing or dynamic environments. This motivates the stronger metric of adaptive regret, or the maximum regret over any continuous sub-interval in time. Existing adaptive regret algorithms suffer from a computational penalty - typically on the order of a multiplicative factor that grows logarithmically in the number of game iterations. In this paper we show how to reduce this computational penalty to be doubly logarithmic in the number of game iterations, and with minimal degradation to the optimal attainable adaptive regret bounds.

Action-modulated midbrain dopamine activity arises from distributed control policies

Animal behavior is driven by multiple brain regions working in parallel with distinct control policies. We present a biologically plausible model of off-policy reinforcement learning in the basal ganglia, which enables learning in such an architecture. The model accounts for action-related modulation of dopamine activity that is not captured by previous models that implement on-policy algorithms. In particular, the model predicts that dopamine activity signals a combination of reward prediction error (as in classic models) and "action surprise," a measure of how unexpected an action is relative to the basal ganglia's current policy. In the presence of the action surprise term, the model implements an approximate form of Q-learning. On benchmark navigation and reaching tasks, we show empirically that this model is capable of learning from data driven completely or in part by other policies (e.g. from other brain regions). By contrast, models without the action surprise term suffer in the presence of additional policies, and are incapable of learning at all from behavior that is completely externally driven. The model provides a computational account for numerous experimental findings about dopamine activity that cannot be explained by classic models of reinforcement learning in the basal ganglia. These include differing levels of action surprise signals in dorsal and ventral striatum, decreasing amounts movement-modulated dopamine activity with practice, and representations of action initiation and kinematics in dopamine activity. It also provides further predictions that can be tested with recordings of striatal dopamine activity.

Offline Policy Optimization with Eligible Actions

Offline policy optimization could have a large impact on many real-world decision-making problems, as online learning may be infeasible in many applications. Importance sampling and its variants are a commonly used type of estimator in offline policy evaluation, and such estimators typically do not require assumptions on the properties and representational capabilities of value function or decision process model function classes. In this paper, we identify an important overfitting phenomenon in optimizing the importance weighted return, in which it may be possible for the learned policy to essentially avoid making aligned decisions for part of the initial state space. We propose an algorithm to avoid this overfitting through a new per-state-neighborhood normalization constraint, and provide a theoretical justification of the proposed algorithm. We also show the limitations of previous attempts to this approach. We test our algorithm in a healthcare-inspired simulator, a logged dataset collected from real hospitals and continuous control tasks. These experiments show the proposed method yields less overfitting and better test performance compared to state-of-the-art batch reinforcement learning algorithms.

Integral Probability Metrics PAC-Bayes Bounds

We present a PAC-Bayes-style generalization bound which enables the replacement of the KL-divergence with a variety of Integral Probability Metrics (IPM). We provide instances of this bound with the IPM being the total variation metric and the Wasserstein distance. A notable feature of the obtained bounds is that they naturally interpolate between classical uniform convergence bounds in the worst case (when the prior and posterior are far away from each other), and preferable bounds in better cases (when the posterior and prior are close). This illustrates the possibility of reinforcing classical generalization bounds with algorithm- and data-dependent components, thus making them more suitable to analyze algorithms that use a large hypothesis space.

FAIR principles for AI models, with a practical application for accelerated high energy diffraction microscopy

A concise and measurable set of FAIR (Findable, Accessible, Interoperable and Reusable) principles for scientific data are transforming the state-of-practice for data management and stewardship, supporting and enabling discovery and innovation. Learning from this initiative, and acknowledging the impact of artificial intelligence (AI) in the practice of science and engineering, we introduce a set of practical, concise and measurable FAIR principles for AI models. We showcase how to create and share FAIR data and AI models within a unified computational framework combining the following elements: the Advanced Photon Source at Argonne National Laboratory, the Materials Data Facility, the Data and Learning Hub for Science, funcX, and the Argonne Leadership Computing Facility (ALCF), in particular the ThetaGPU supercomputer and the SambaNova DataScale system at the ALCF AI-Testbed. We describe how this domain-agnostic computational framework may be harnessed to enable autonomous AI-driven discovery.

A Temporal Fusion Transformer for Long-term Explainable Prediction of Emergency Department Overcrowding

Emergency Departments (EDs) are a fundamental element of the Portuguese National Health Service, serving as an entry point for users with diverse and very serious medical problems. Due to the inherent characteristics of the ED; forecasting the number of patients using the services is particularly challenging. And a mismatch between the affluence and the number of medical professionals can lead to a decrease in the quality of the services provided and create problems that have repercussions for the entire hospital, with the requisition of health care workers from other departments and the postponement of surgeries. ED overcrowding is driven, in part, by non-urgent patients, that resort to emergency services despite not having a medical emergency and which represent almost half of the total number of daily patients. This paper describes a novel deep learning architecture, the Temporal Fusion Transformer, that uses calendar and time-series covariates to forecast prediction intervals and point predictions for a 4 week period. We have concluded that patient volume can be forecasted with a Mean Absolute Percentage Error (MAPE) of 5.90% for Portugal's Health Regional Areas (HRA) and a Root Mean Squared Error (RMSE) of 84.4102 people/day. The paper shows empirical evidence supporting the use of a multivariate approach with static and time-series covariates while surpassing other models commonly found in the literature.

On Leave-One-Out Conditional Mutual Information For Generalization

We derive information theoretic generalization bounds for supervised learning algorithms based on a new measure of leave-one-out conditional mutual information (loo-CMI). Contrary to other CMI bounds, which are black-box bounds that do not exploit the structure of the problem and may be hard to evaluate in practice, our loo-CMI bounds can be computed easily and can be interpreted in connection to other notions such as classical leave-one-out cross-validation, stability of the optimization algorithm, and the geometry of the loss-landscape. It applies both to the output of training algorithms as well as their predictions. We empirically validate the quality of the bound by evaluating its predicted generalization gap in scenarios for deep learning. In particular, our bounds are non-vacuous on large-scale image-classification tasks.

Video + CLIP Baseline for Ego4D Long-term Action Anticipation

In this report, we introduce our adaptation of image-text models for long-term action anticipation. Our Video + CLIP framework makes use of a large-scale pre-trained paired image-text model: CLIP and a video encoder Slowfast network. The CLIP embedding provides fine-grained understanding of objects relevant for an action whereas the slowfast network is responsible for modeling temporal information within a video clip of few frames. We show that the features obtained from both encoders are complementary to each other, thus outperforming the baseline on Ego4D for the task of long-term action anticipation. Our code is available at github.com/srijandas07/clip_baseline_LTA_Ego4d.

How can spherical CNNs benefit ML-based diffusion MRI parameter estimation?

This paper demonstrates spherical convolutional neural networks (S-CNN) offer distinct advantages over conventional fully-connected networks (FCN) at estimating scalar parameters of tissue microstructure from diffusion MRI (dMRI). Such microstructure parameters are valuable for identifying pathology and quantifying its extent. However, current clinical practice commonly acquires dMRI data consisting of only 6 diffusion weighted images (DWIs), limiting the accuracy and precision of estimated microstructure indices. Machine learning (ML) has been proposed to address this challenge. However, existing ML-based methods are not robust to differing dMRI gradient sampling schemes, nor are they rotation equivariant. Lack of robustness to sampling schemes requires a new network to be trained for each scheme, complicating the analysis of data from multiple sources. A possible consequence of the lack of rotational equivariance is that the training dataset must contain a diverse range of microstucture orientations. Here, we show spherical CNNs represent a compelling alternative that is robust to new sampling schemes as well as offering rotational equivariance. We show the latter can be leveraged to decrease the number of training datapoints required.

Ultra-low latency recurrent neural network inference on FPGAs for physics applications with hls4ml

Recurrent neural networks have been shown to be effective architectures for many tasks in high energy physics, and thus have been widely adopted. Their use in low-latency environments has, however, been limited as a result of the difficulties of implementing recurrent architectures on field-programmable gate arrays (FPGAs). In this paper we present an implementation of two types of recurrent neural network layers -- long short-term memory and gated recurrent unit -- within the hls4ml framework. We demonstrate that our implementation is capable of producing effective designs for both small and large models, and can be customized to meet specific design requirements for inference latencies and FPGA resources. We show the performance and synthesized designs for multiple neural networks, many of which are trained specifically for jet identification tasks at the CERN Large Hadron Collider.

Learning to correct spectral methods for simulating turbulent flows

Despite their ubiquity throughout science and engineering, only a handful of partial differential equations (PDEs) have analytical, or closed-form solutions. This motivates a vast amount of classical work on numerical simulation of PDEs and more recently, a whirlwind of research into data-driven techniques leveraging machine learning (ML). A recent line of work indicates that a hybrid of classical numerical techniques with machine learning can offer significant improvements over either approach alone. In this work, we show that the choice of the numerical scheme is crucial when incorporating physics-based priors. We build upon Fourier-based spectral methods, which are considerably more efficient than other numerical schemes for simulating PDEs with smooth and periodic solutions. Specifically, we develop ML-augmented spectral solvers for three model PDEs of fluid dynamics, which improve upon the accuracy of standard spectral solvers at the same resolution. We also demonstrate a handful of key design principles for combining machine learning and numerical methods for solving PDEs.

FitHuBERT: Going Thinner and Deeper for Knowledge Distillation of Speech Self-Supervised Learning

Large-scale speech self-supervised learning (SSL) has emerged to the main field of speech processing, however, the problem of computational cost arising from its vast size makes a high entry barrier to academia. In addition, existing distillation techniques of speech SSL models compress the model by reducing layers, which induces performance degradation in linguistic pattern recognition tasks such as phoneme recognition (PR). In this paper, we propose FitHuBERT, which makes thinner in dimension throughout almost all model components and deeper in layer compared to prior speech SSL distillation works. Moreover, we employ a time-reduction layer to speed up inference time and propose a method of hint-based distillation for less performance degradation. Our method reduces the model to 23.8% in size and 35.9% in inference time compared to HuBERT. Also, we achieve 12.1% word error rate and 13.3% phoneme error rate on the SUPERB benchmark which is superior than prior work.

Evaluating the Explainers: Black-Box Explainable Machine Learning for Student Success Prediction in MOOCs

Neural networks are ubiquitous in applied machine learning for education. Their pervasive success in predictive performance comes alongside a severe weakness, the lack of explainability of their decisions, especially relevant in human-centric fields. We implement five state-of-the-art methodologies for explaining black-box machine learning models (LIME, PermutationSHAP, KernelSHAP, DiCE, CEM) and examine the strengths of each approach on the downstream task of student performance prediction for five massive open online courses. Our experiments demonstrate that the families of explainers do not agree with each other on feature importance for the same Bidirectional LSTM models with the same representative set of students. We use Principal Component Analysis, Jensen-Shannon distance, and Spearman's rank-order correlation to quantitatively cross-examine explanations across methods and courses. Furthermore, we validate explainer performance across curriculum-based prerequisite relationships. Our results come to the concerning conclusion that the choice of explainer is an important decision and is in fact paramount to the interpretation of the predictive results, even more so than the course the model is trained on. Source code and models are released at http://github.com/epfl-ml4ed/evaluating-explainers.

The "AI+R"-tree: An Instance-optimized R-tree

The emerging class of instance-optimized systems has shown potential to achieve high performance by specializing to a specific data and query workloads. Particularly, Machine Learning (ML) techniques have been applied successfully to build various instance-optimized components (e.g., learned indexes). This paper investigates to leverage ML techniques to enhance the performance of spatial indexes, particularly the R-tree, for a given data and query workloads. As the areas covered by the R-tree index nodes overlap in space, upon searching for a specific point in space, multiple paths from root to leaf may potentially be explored. In the worst case, the entire R-tree could be searched. In this paper, we define and use the overlap ratio to quantify the degree of extraneous leaf node accesses required by a range query. The goal is to enhance the query performance of a traditional R-tree for high-overlap range queries as they tend to incur long running-times. We introduce a new AI-tree that transforms the search operation of an R-tree into a multi-label classification task to exclude the extraneous leaf node accesses. Then, we augment a traditional R-tree to the AI-tree to form a hybrid "AI+R"-tree. The "AI+R"-tree can automatically differentiate between the high- and low-overlap queries using a learned model. Thus, the "AI+R"-tree processes high-overlap queries using the AI-tree, and the low-overlap queries using the R-tree. Experiments on real datasets demonstrate that the "AI+R"-tree can enhance the query performance over a traditional R-tree by up to 500%.

Masked Autoencoders for Self-Supervised Learning on Automotive Point Clouds

Masked autoencoding has become a successful pre-training paradigm for Transformer models for text, images, and recently, point clouds. Raw automotive datasets are a suitable candidate for self-supervised pre-training as they generally are cheap to collect compared to annotations for tasks like 3D object detection (OD). However, development of masked autoencoders for point clouds has focused solely on synthetic and indoor data. Consequently, existing methods have tailored their representations and models toward point clouds which are small, dense and have homogeneous point density. In this work, we study masked autoencoding for point clouds in an automotive setting, which are sparse and for which the point density can vary drastically among objects in the same scene. To this end, we propose Voxel-MAE, a simple masked autoencoding pre-training scheme designed for voxel representations. We pre-train the backbone of a Transformer-based 3D object detector to reconstruct masked voxels and to distinguish between empty and non-empty voxels. Our method improves the 3D OD performance by 1.75 mAP points and 1.05 NDS on the challenging nuScenes dataset. Compared to existing self-supervised methods for automotive data, Voxel-MAE displays up to $2\times$ performance increase. Further, we show that by pre-training with Voxel-MAE, we require only 40% of the annotated data to outperform a randomly initialized equivalent. Code will be released.