Machine Learning (cs.LG)

Abstract: In this work, we propose a novel activation mechanism aimed at establishing layer-level activation (LayerAct) functions. These functions are designed to be more noise-robust compared to traditional element-level activation functions by reducing the layer-level fluctuation of the activation outputs due to shift in inputs. Moreover, the LayerAct functions achieve a zero-like mean activation output without restricting the activation output space. We present an analysis and experiments demonstrating that LayerAct functions exhibit superior noise-robustness compared to element-level activation functions, and empirically show that these functions have a zero-like mean activation. Experimental results on three benchmark image classification tasks show that LayerAct functions excel in handling noisy image datasets, outperforming element-level activation functions, while the performance on clean datasets is also superior in most cases.

Abstract: With the recent progress in sports analytics, deep learning approaches have demonstrated the effectiveness of mining insights into players' tactics for improving performance quality and fan engagement. This is attributed to the availability of public ground-truth datasets. While there are a few available datasets for turn-based sports for action detection, these datasets severely lack structured source data and stroke-level records since these require high-cost labeling efforts from domain experts and are hard to detect using automatic techniques. Consequently, the development of artificial intelligence approaches is significantly hindered when existing models are applied to more challenging structured turn-based sequences. In this paper, we present ShuttleSet, the largest publicly-available badminton singles dataset with annotated stroke-level records. It contains 104 sets, 3,685 rallies, and 36,492 strokes in 44 matches between 2018 and 2021 with 27 top-ranking men's singles and women's singles players. ShuttleSet is manually annotated with a computer-aided labeling tool to increase the labeling efficiency and effectiveness of selecting the shot type with a choice of 18 distinct classes, the corresponding hitting locations, and the locations of both players at each stroke. In the experiments, we provide multiple benchmarks (i.e., stroke influence, stroke forecasting, and movement forecasting) with baselines to illustrate the practicability of using ShuttleSet for turn-based analytics, which is expected to stimulate both academic and sports communities. Over the past two years, a visualization platform has been deployed to illustrate the variability of analysis cases from ShuttleSet for coaches to delve into players' tactical preferences with human-interactive interfaces, which was also used by national badminton teams during multiple international high-ranking matches.

Abstract: While deep learning models have shown remarkable performance in various tasks, they are susceptible to learning non-generalizable spurious features rather than the core features that are genuinely correlated to the true label. In this paper, beyond existing analyses of linear models, we theoretically examine the learning process of a two-layer nonlinear convolutional neural network in the presence of spurious features. Our analysis suggests that imbalanced data groups and easily learnable spurious features can lead to the dominance of spurious features during the learning process. In light of this, we propose a new training algorithm called PDE that efficiently enhances the model's robustness for a better worst-group performance. PDE begins with a group-balanced subset of training data and progressively expands it to facilitate the learning of the core features. Experiments on synthetic and real-world benchmark datasets confirm the superior performance of our method on models such as ResNets and Transformers. On average, our method achieves a 2.8% improvement in worst-group accuracy compared with the state-of-the-art method, while enjoying up to 10x faster training efficiency.

Abstract: We propose a new algorithm for the problem of recovering data that adheres to multiple, heterogeneous low-dimensional structures from linear observations. Focusing on data matrices that are simultaneously row-sparse and low-rank, we propose and analyze an iteratively reweighted least squares (IRLS) algorithm that is able to leverage both structures. In particular, it optimizes a combination of non-convex surrogates for row-sparsity and rank, a balancing of which is built into the algorithm. We prove locally quadratic convergence of the iterates to a simultaneously structured data matrix in a regime of minimal sample complexity (up to constants and a logarithmic factor), which is known to be impossible for a combination of convex surrogates. In experiments, we show that the IRLS method exhibits favorable empirical convergence, identifying simultaneously row-sparse and low-rank matrices from fewer measurements than state-of-the-art methods.

Abstract: Errors of machine learning models are costly, especially in safety-critical domains such as healthcare, where such mistakes can prevent the deployment of machine learning altogether. In these settings, conservative models -- models which can defer to human judgment when they are likely to make an error -- may offer a solution. However, detecting unusual or difficult examples is notably challenging, as it is impossible to anticipate all potential inputs at test time. To address this issue, prior work has proposed to minimize the model's confidence on an auxiliary pseudo-OOD dataset. We theoretically analyze the effect of confidence minimization and show that the choice of auxiliary dataset is critical. Specifically, if the auxiliary dataset includes samples from the OOD region of interest, confidence minimization provably separates ID and OOD inputs by predictive confidence. Taking inspiration from this result, we present data-driven confidence minimization (DCM), which minimizes confidence on an uncertainty dataset containing examples that the model is likely to misclassify at test time. Our experiments show that DCM consistently outperforms state-of-the-art OOD detection methods on 8 ID-OOD dataset pairs, reducing FPR (at TPR 95%) by 6.3% and 58.1% on CIFAR-10 and CIFAR-100, and outperforms existing selective classification approaches on 4 datasets in conditions of distribution shift.

Abstract: Although graph neural networks (GNNs) have achieved impressive achievements in graph classification, they often need abundant task-specific labels, which could be extensively costly to acquire. A credible solution is to explore additional labeled graphs to enhance unsupervised learning on the target domain. However, how to apply GNNs to domain adaptation remains unsolved owing to the insufficient exploration of graph topology and the significant domain discrepancy. In this paper, we propose \underline{Co}upled \underline{Co}ntrastive Graph Representation Learning (\method{}), which extracts the topological information from coupled learning branches and reduces the domain discrepancy with coupled contrastive learning. \method{} contains a graph convolutional network branch and a hierarchical graph kernel network branch, which explore graph topology in implicit and explicit manners. Besides, we incorporate coupled branches into a holistic multi-view contrastive learning framework, which not only incorporates graph representations learned from complementary views for enhanced understanding, but also encourages the similarity between cross-domain example pairs with the same semantics for domain alignment. Extensive experiments on various popular datasets show that \method{} outperforms these competing baselines by 5.7\% to 21.0\% generally.

Abstract: Research has shown that deep networks tend to be overly optimistic about their predictions, leading to an underestimation of prediction errors. Due to the limited nature of data, existing studies have proposed various methods based on model prediction probabilities to bin the data and evaluate calibration error. We propose a more generalized definition of calibration error called Partitioned Calibration Error (PCE), revealing that the key difference among these calibration error metrics lies in how the data space is partitioned. We put forth an intuitive proposition that an accurate model should be calibrated across any partition, suggesting that the input space partitioning can extend beyond just the partitioning of prediction probabilities, and include partitions directly related to the input. Through semantic-related partitioning functions, we demonstrate that the relationship between model accuracy and calibration lies in the granularity of the partitioning function. This highlights the importance of partitioning criteria for training a calibrated and accurate model. To validate the aforementioned analysis, we propose a method that involves jointly learning a semantic aware grouping function based on deep model features and logits to partition the data space into subsets. Subsequently, a separate calibration function is learned for each subset. Experimental results demonstrate that our approach achieves significant performance improvements across multiple datasets and network architectures, thus highlighting the importance of the partitioning function for calibration.

Abstract: Minimizing response times is crucial for emergency medical services to reduce patients' waiting times and to increase their survival rates. Many models exist to optimize operational tasks such as ambulance allocation and dispatching. Including accurate demand forecasts in such models can improve operational decision-making. Against this background, we present a novel convolutional neural network (CNN) architecture that transforms time series data into heatmaps to predict ambulance demand. Applying such predictions requires incorporating external features that influence ambulance demands. We contribute to the existing literature by providing a flexible, generic CNN architecture, allowing for the inclusion of external features with varying dimensions. Additionally, we provide a feature selection and hyperparameter optimization framework utilizing Bayesian optimization. We integrate historical ambulance demand and external information such as weather, events, holidays, and time. To show the superiority of the developed CNN architecture over existing approaches, we conduct a case study for Seattle's 911 call data and include external information. We show that the developed CNN architecture outperforms existing state-of-the-art methods and industry practice by more than 9%.

Abstract: Neural Network designs are quite diverse, from VGG-style to ResNet-style, and from Convolutional Neural Networks to Transformers. Towards the design of efficient accelerators, many works have adopted a dataflow-based, inter-layer pipelined architecture, with a customised hardware towards each layer, achieving ultra high throughput and low latency. The deployment of neural networks to such dataflow architecture accelerators is usually hindered by the available on-chip memory as it is desirable to preload the weights of neural networks on-chip to maximise the system performance. To address this, networks are usually compressed before the deployment through methods such as pruning, quantization and tensor decomposition. In this paper, a framework for mapping CNNs onto FPGAs based on a novel tensor decomposition method called Mixed-TD is proposed. The proposed method applies layer-specific Singular Value Decomposition (SVD) and Canonical Polyadic Decomposition (CPD) in a mixed manner, achieving 1.73x to 10.29x throughput per DSP to state-of-the-art CNNs. Our work is open-sourced: https://github.com/Yu-Zhewen/Mixed-TD

Abstract: Recent neural architecture search (NAS) frameworks have been successful in finding optimal architectures for given conditions (e.g., performance or latency). However, they search for optimal architectures in terms of their performance on clean images only, while robustness against various types of perturbations or corruptions is crucial in practice. Although there exist several robust NAS frameworks that tackle this issue by integrating adversarial training into one-shot NAS, however, they are limited in that they only consider robustness against adversarial attacks and require significant computational resources to discover optimal architectures for a single task, which makes them impractical in real-world scenarios. To address these challenges, we propose a novel lightweight robust zero-cost proxy that considers the consistency across features, parameters, and gradients of both clean and perturbed images at the initialization state. Our approach facilitates an efficient and rapid search for neural architectures capable of learning generalizable features that exhibit robustness across diverse perturbations. The experimental results demonstrate that our proxy can rapidly and efficiently search for neural architectures that are consistently robust against various perturbations on multiple benchmark datasets and diverse search spaces, largely outperforming existing clean zero-shot NAS and robust NAS with reduced search cost.

Abstract: As Transformer-based models have achieved impressive performance on various time series tasks, Long-Term Series Forecasting (LTSF) tasks have also received extensive attention in recent years. However, due to the inherent computational complexity and long sequences demanding of Transformer-based methods, its application on LTSF tasks still has two major issues that need to be further investigated: 1) Whether the sparse attention mechanism designed by these methods actually reduce the running time on real devices; 2) Whether these models need extra long input sequences to guarantee their performance? The answers given in this paper are negative. Therefore, to better copy with these two issues, we design a lightweight Period-Attention mechanism (Periodformer), which renovates the aggregation of long-term subseries via explicit periodicity and short-term subseries via built-in proximity. Meanwhile, a gating mechanism is embedded into Periodformer to regulate the influence of the attention module on the prediction results. Furthermore, to take full advantage of GPUs for fast hyperparameter optimization (e.g., finding the suitable input length), a Multi-GPU Asynchronous parallel algorithm based on Bayesian Optimization (MABO) is presented. MABO allocates a process to each GPU via a queue mechanism, and then creates multiple trials at a time for asynchronous parallel search, which greatly reduces the search time. Compared with the state-of-the-art methods, the prediction error of Periodformer reduced by 13% and 26% for multivariate and univariate forecasting, respectively. In addition, MABO reduces the average search time by 46% while finding better hyperparameters. As a conclusion, this paper indicates that LTSF may not need complex attention and extra long input sequences. The source code will be open source on Github.

Abstract: In this paper, we introduce a novel semantic generative communication (SGC) framework, where generative users leverage text-to-image (T2I) generators to create images locally from downloaded text prompts, while non-generative users directly download images from a base station (BS). Although generative users help reduce downlink transmission energy at the BS, they consume additional energy for image generation and for uploading their generator state information (GSI). We formulate the problem of minimizing the total energy consumption of the BS and the users, and devise a generative user selection algorithm. Simulation results corroborate that our proposed algorithm reduces total energy by up to 54% compared to a baseline with all non-generative users.

Abstract: Recently, denoising diffusion models have led to significant breakthroughs in the generation of images, audio and text. However, it is still an open question on how to adapt their strong modeling ability to model time series. In this paper, we propose TimeDiff, a non-autoregressive diffusion model that achieves high-quality time series prediction with the introduction of two novel conditioning mechanisms: future mixup and autoregressive initialization. Similar to teacher forcing, future mixup allows parts of the ground-truth future predictions for conditioning, while autoregressive initialization helps better initialize the model with basic time series patterns such as short-term trends. Extensive experiments are performed on nine real-world datasets. Results show that TimeDiff consistently outperforms existing time series diffusion models, and also achieves the best overall performance across a variety of the existing strong baselines (including transformers and FiLM).

Abstract: Learning with noisy labels is an important topic for scalable training in many real-world scenarios. However, few previous research considers this problem in the online setting, where the arrival of data is streaming. In this paper, we propose a novel gradient-based approach to enable the detection of noisy labels for the online learning of model parameters, named Online Gradient-based Robust Selection (OGRS). In contrast to the previous sample selection approach for the offline training that requires the estimation of a clean ratio of the dataset before each epoch of training, OGRS can automatically select clean samples by steps of gradient update from datasets with varying clean ratios without changing the parameter setting. During the training process, the OGRS method selects clean samples at each iteration and feeds the selected sample to incrementally update the model parameters. We provide a detailed theoretical analysis to demonstrate data selection process is converging to the low-loss region of the sample space, by introducing and proving the sub-linear local Lagrangian regret of the non-convex constrained optimization problem. Experimental results show that it outperforms state-of-the-art methods in different settings.

Abstract: Tabular data is often hidden in text, particularly in medical diagnostic reports. Traditional machine learning (ML) models designed to work with tabular data, cannot effectively process information in such form. On the other hand, large language models (LLMs) which excel at textual tasks, are probably not the best tool for modeling tabular data. Therefore, we propose a novel, simple, and effective methodology for extracting structured tabular data from textual medical reports, called TEMED-LLM. Drawing upon the reasoning capabilities of LLMs, TEMED-LLM goes beyond traditional extraction techniques, accurately inferring tabular features, even when their names are not explicitly mentioned in the text. This is achieved by combining domain-specific reasoning guidelines with a proposed data validation and reasoning correction feedback loop. By applying interpretable ML models such as decision trees and logistic regression over the extracted and validated data, we obtain end-to-end interpretable predictions. We demonstrate that our approach significantly outperforms state-of-the-art text classification models in medical diagnostics. Given its predictive performance, simplicity, and interpretability, TEMED-LLM underscores the potential of leveraging LLMs to improve the performance and trustworthiness of ML models in medical applications.

Abstract: Deep Learning models are a standard solution for sensor-based Human Activity Recognition (HAR), but their deployment is often limited by labeled data scarcity and models' opacity. Neuro-Symbolic AI (NeSy) provides an interesting research direction to mitigate these issues by infusing knowledge about context information into HAR deep learning classifiers. However, existing NeSy methods for context-aware HAR require computationally expensive symbolic reasoners during classification, making them less suitable for deployment on resource-constrained devices (e.g., mobile devices). Additionally, NeSy approaches for context-aware HAR have never been evaluated on in-the-wild datasets, and their generalization capabilities in real-world scenarios are questionable. In this work, we propose a novel approach based on a semantic loss function that infuses knowledge constraints in the HAR model during the training phase, avoiding symbolic reasoning during classification. Our results on scripted and in-the-wild datasets show the impact of different semantic loss functions in outperforming a purely data-driven model. We also compare our solution with existing NeSy methods and analyze each approach's strengths and weaknesses. Our semantic loss remains the only NeSy solution that can be deployed as a single DNN without the need for symbolic reasoning modules, reaching recognition rates close (and better in some cases) to existing approaches.

Abstract: The need to execute Deep Neural Networks (DNNs) at low latency and low power at the edge has spurred the development of new heterogeneous Systems-on-Chips (SoCs) encapsulating a diverse set of hardware accelerators. How to optimally map a DNN onto such multi-accelerator systems is an open problem. We propose ODiMO, a hardware-aware tool that performs a fine-grain mapping across different accelerators on-chip, splitting individual layers and executing them in parallel, to reduce inference energy consumption or latency, while taking into account each accelerator's quantization precision to maintain accuracy. Pareto-optimal networks in the accuracy vs. energy or latency space are pursued for three popular dataset/DNN pairs, and deployed on the DIANA heterogeneous ultra-low power edge AI SoC. We show that ODiMO reduces energy/latency by up to 33%/31% with limited accuracy drop (-0.53%/-0.32%) compared to manual heuristic mappings.

Abstract: Visual Prompt Tuning (VPT) is an effective tuning method for adapting pretrained Vision Transformers (ViTs) to downstream tasks. It leverages extra learnable tokens, known as prompts, which steer the frozen pretrained ViTs. Although VPT has demonstrated its applicability with supervised vision transformers, it often underperforms with self-supervised ones. Through empirical observations, we deduce that the effectiveness of VPT hinges largely on the ViT blocks with which the prompt tokens interact. Specifically, VPT shows improved performance on image classification tasks for MAE and MoCo v3 when the prompt tokens are inserted into later blocks rather than the first block. These observations suggest that there exists an optimal location of blocks for the insertion of prompt tokens. Unfortunately, identifying the optimal blocks for prompts within each self-supervised ViT for diverse future scenarios is a costly process. To mitigate this problem, we propose a simple yet effective method that learns a gate for each ViT block to adjust its intervention into the prompt tokens. With our method, prompt tokens are selectively influenced by blocks that require steering for task adaptation. Our method outperforms VPT variants in FGVC and VTAB image classification and ADE20K semantic segmentation. The code is available at https://github.com/ryongithub/GatedPromptTuning.

Abstract: Accurately measuring discrimination is crucial to faithfully assessing fairness of trained machine learning (ML) models. Any bias in measuring discrimination leads to either amplification or underestimation of the existing disparity. Several sources of bias exist and it is assumed that bias resulting from machine learning is born equally by different groups (e.g. females vs males, whites vs blacks, etc.). If, however, bias is born differently by different groups, it may exacerbate discrimination against specific sub-populations. Sampling bias, is inconsistently used in the literature to describe bias due to the sampling procedure. In this paper, we attempt to disambiguate this term by introducing clearly defined variants of sampling bias, namely, sample size bias (SSB) and underrepresentation bias (URB). We show also how discrimination can be decomposed into variance, bias, and noise. Finally, we challenge the commonly accepted mitigation approach that discrimination can be addressed by collecting more samples of the underrepresented group.

Abstract: In this work, we present a method to add perturbations to the code descriptions, i.e., new inputs in natural language (NL) from well-intentioned developers, in the context of security-oriented code, and analyze how and to what extent perturbations affect the performance of AI offensive code generators. Our experiments show that the performance of the code generators is highly affected by perturbations in the NL descriptions. To enhance the robustness of the code generators, we use the method to perform data augmentation, i.e., to increase the variability and diversity of the training data, proving its effectiveness against both perturbed and non-perturbed code descriptions.

Abstract: We introduce a novel and general loss function, called Symmetric Contrastive (Sy-CON) loss, for effective continual self-supervised learning (CSSL). We first argue that the conventional loss form of continual learning which consists of single task-specific loss (for plasticity) and a regularizer (for stability) may not be ideal for contrastive loss based CSSL that focus on representation learning. Our reasoning is that, in contrastive learning based methods, the task-specific loss would suffer from decreasing diversity of negative samples and the regularizer may hinder learning new distinctive representations. To that end, we propose Sy-CON that consists of two losses (one for plasticity and the other for stability) with symmetric dependence on current and past models' negative sample embeddings. We argue our model can naturally find good trade-off between the plasticity and stability without any explicit hyperparameter tuning. We validate the effectiveness of our approach through extensive experiments, demonstrating that MoCo-based implementation of Sy-CON loss achieves superior performance compared to other state-of-the-art CSSL methods.

Abstract: Graphs are widely used to encapsulate a variety of data formats, but real-world networks often involve complex node relations beyond only being pairwise. While hypergraphs and hierarchical graphs have been developed and employed to account for the complex node relations, they cannot fully represent these complexities in practice. Additionally, though many Graph Neural Networks (GNNs) have been proposed for representation learning on higher-order graphs, they are usually only evaluated on simple graph datasets. Therefore, there is a need for a unified modelling of higher-order graphs, and a collection of comprehensive datasets with an accessible evaluation framework to fully understand the performance of these algorithms on complex graphs. In this paper, we introduce the concept of hybrid graphs, a unified definition for higher-order graphs, and present the Hybrid Graph Benchmark (HGB). HGB contains 23 real-world hybrid graph datasets across various domains such as biology, social media, and e-commerce. Furthermore, we provide an extensible evaluation framework and a supporting codebase to facilitate the training and evaluation of GNNs on HGB. Our empirical study of existing GNNs on HGB reveals various research opportunities and gaps, including (1) evaluating the actual performance improvement of hypergraph GNNs over simple graph GNNs; (2) comparing the impact of different sampling strategies on hybrid graph learning methods; and (3) exploring ways to integrate simple graph and hypergraph information. We make our source code and full datasets publicly available at https://zehui127.github.io/hybrid-graph-benchmark/.

Abstract: Medical applications of machine learning (ML) have experienced a surge in popularity in recent years. The intensive care unit (ICU) is a natural habitat for ML given the abundance of available data from electronic health records. Models have been proposed to address numerous ICU prediction tasks like the early detection of complications. While authors frequently report state-of-the-art performance, it is challenging to verify claims of superiority. Datasets and code are not always published, and cohort definitions, preprocessing pipelines, and training setups are difficult to reproduce. This work introduces Yet Another ICU Benchmark (YAIB), a modular framework that allows researchers to define reproducible and comparable clinical ML experiments; we offer an end-to-end solution from cohort definition to model evaluation. The framework natively supports most open-access ICU datasets (MIMIC III/IV, eICU, HiRID, AUMCdb) and is easily adaptable to future ICU datasets. Combined with a transparent preprocessing pipeline and extensible training code for multiple ML and deep learning models, YAIB enables unified model development. Our benchmark comes with five predefined established prediction tasks (mortality, acute kidney injury, sepsis, kidney function, and length of stay) developed in collaboration with clinicians. Adding further tasks is straightforward by design. Using YAIB, we demonstrate that the choice of dataset, cohort definition, and preprocessing have a major impact on the prediction performance - often more so than model class - indicating an urgent need for YAIB as a holistic benchmarking tool. We provide our work to the clinical ML community to accelerate method development and enable real-world clinical implementations. Software Repository: https://github.com/rvandewater/YAIB.

Abstract: Generative design is an increasingly important tool in the industrial world. It allows the designers and engineers to easily explore vast ranges of design options, providing a cheaper and faster alternative to the trial and failure approaches. Thanks to the flexibility they offer, Deep Generative Models are gaining popularity amongst Generative Design technologies. However, developing and evaluating these models can be challenging. The field lacks accessible benchmarks, in order to evaluate and compare objectively different Deep Generative Models architectures. Moreover, vanilla Deep Generative Models appear to be unable to accurately generate multi-components industrial systems that are controlled by latent design constraints. To address these challenges, we propose an industry-inspired use case that incorporates actual industrial system characteristics. This use case can be quickly generated and used as a benchmark. We propose a Meta-VAE capable of producing multi-component industrial systems and showcase its application on the proposed use case.

Abstract: Given the increasing volume and quality of genomics data, extracting new insights requires interpretable machine-learning models. This work presents Genomic Interpreter: a novel architecture for genomic assay prediction. This model outperforms the state-of-the-art models for genomic assay prediction tasks. Our model can identify hierarchical dependencies in genomic sites. This is achieved through the integration of 1D-Swin, a novel Transformer-based block designed by us for modelling long-range hierarchical data. Evaluated on a dataset containing 38,171 DNA segments of 17K base pairs, Genomic Interpreter demonstrates superior performance in chromatin accessibility and gene expression prediction and unmasks the underlying `syntax' of gene regulation.

Abstract: We consider the problem of optimizing a grey-box objective function, i.e., nested function composed of both black-box and white-box functions. A general formulation for such grey-box problems is given, which covers the existing grey-box optimization formulations as special cases. We then design an optimism-driven algorithm to solve it. Under certain regularity assumptions, our algorithm achieves similar regret bound as that for the standard black-box Bayesian optimization algorithm, up to a constant multiplicative term depending on the Lipschitz constants of the functions considered. We further extend our method to the constrained case and discuss several special cases. For the commonly used kernel functions, the regret bounds allow us to derive a convergence rate to the optimal solution. Experimental results show that our grey-box optimization method empirically improves the speed of finding the global optimal solution significantly, as compared to the standard black-box optimization algorithm.

Abstract: Recently, sequence learning methods have been applied to the problem of off-policy Reinforcement Learning, including the seminal work on Decision Transformers, which employs transformers for this task. Since transformers are parameter-heavy, cannot benefit from history longer than a fixed window size, and are not computed using recurrence, we set out to investigate the suitability of the S4 family of models, which are based on state-space layers and have been shown to outperform transformers, especially in modeling long-range dependencies. In this work we present two main algorithms: (i) an off-policy training procedure that works with trajectories, while still maintaining the training efficiency of the S4 model. (ii) An on-policy training procedure that is trained in a recurrent manner, benefits from long-range dependencies, and is based on a novel stable actor-critic mechanism. Our results indicate that our method outperforms multiple variants of decision transformers, as well as the other baseline methods on most tasks, while reducing the latency, number of parameters, and training time by several orders of magnitude, making our approach more suitable for real-world RL.

Abstract: Federated Machine Learning (FL) has received considerable attention in recent years. FL benchmarks are predominantly explored in either simulated systems or data center environments, neglecting the setups of real-world systems, which are often closely linked to edge computing. We close this research gap by introducing FLEdge, a benchmark targeting FL workloads in edge computing systems. We systematically study hardware heterogeneity, energy efficiency during training, and the effect of various differential privacy levels on training in FL systems. To make this benchmark applicable to real-world scenarios, we evaluate the impact of client dropouts on state-of-the-art FL strategies with failure rates as high as 50%. FLEdge provides new insights, such as that training state-of-the-art FL workloads on older GPU-accelerated embedded devices is up to 3x more energy efficient than on modern server-grade GPUs.

Abstract: The state of the art of many learning tasks, e.g., image classification, is advanced by collecting larger datasets and then training larger models on them. As the outcome, the increasing computational cost is becoming unaffordable. In this paper, we investigate how to prune the large-scale datasets, and thus produce an informative subset for training sophisticated deep models with negligible performance drop. We propose a simple yet effective dataset pruning method by exploring both the prediction uncertainty and training dynamics. To our knowledge, this is the first work to study dataset pruning on large-scale datasets, i.e., ImageNet-1K and ImageNet-21K, and advanced models, i.e., Swin Transformer and ConvNeXt. Extensive experimental results indicate that our method outperforms the state of the art and achieves 75% lossless compression ratio on both ImageNet-1K and ImageNet-21K. The code and pruned datasets are available at https://github.com/BAAI-DCAI/Dataset-Pruning.

Abstract: Few-shot meta-learning presents a challenge for gradient descent optimization due to the limited number of training samples per task. To address this issue, we propose an episodic memory optimization for meta-learning, we call \emph{EMO}, which is inspired by the human ability to recall past learning experiences from the brain's memory. EMO retains the gradient history of past experienced tasks in external memory, enabling few-shot learning in a memory-augmented way. By learning to retain and recall the learning process of past training tasks, EMO nudges parameter updates in the right direction, even when the gradients provided by a limited number of examples are uninformative. We prove theoretically that our algorithm converges for smooth, strongly convex objectives. EMO is generic, flexible, and model-agnostic, making it a simple plug-and-play optimizer that can be seamlessly embedded into existing optimization-based few-shot meta-learning approaches. Empirical results show that EMO scales well with most few-shot classification benchmarks and improves the performance of optimization-based meta-learning methods, resulting in accelerated convergence.

Abstract: Understanding the characteristics of neural networks is important but difficult due to their complex structures and behaviors. Some previous work proposes to transform neural networks into equivalent Boolean expressions and apply verification techniques for characteristics of interest. This approach is promising since rich results of verification techniques for circuits and other Boolean expressions can be readily applied. The bottleneck is the time complexity of the transformation. More precisely, (i) each neuron of the network, i.e., a linear threshold function, is converted to a Binary Decision Diagram (BDD), and (ii) they are further combined into some final form, such as Boolean circuits. For a linear threshold function with $n$ variables, an existing method takes $O(n2^{\frac{n}{2}})$ time to construct an ordered BDD of size $O(2^{\frac{n}{2}})$ consistent with some variable ordering. However, it is non-trivial to choose a variable ordering producing a small BDD among $n!$ candidates. We propose a method to convert a linear threshold function to a specific form of a BDD based on the boosting approach in the machine learning literature. Our method takes $O(2^n \text{poly}(1/\rho))$ time and outputs BDD of size $O(\frac{n^2}{\rho^4}\ln{\frac{1}{\rho}})$, where $\rho$ is the margin of some consistent linear threshold function. Our method does not need to search for good variable orderings and produces a smaller expression when the margin of the linear threshold function is large. More precisely, our method is based on our new boosting algorithm, which is of independent interest. We also propose a method to combine them into the final Boolean expression representing the neural network.

Abstract: Machine learning head models (MLHMs) are developed to estimate brain deformation for early detection of traumatic brain injury (TBI). However, the overfitting to simulated impacts and the lack of generalizability caused by distributional shift of different head impact datasets hinders the broad clinical applications of current MLHMs. We propose brain deformation estimators that integrates unsupervised domain adaptation with a deep neural network to predict whole-brain maximum principal strain (MPS) and MPS rate (MPSR). With 12,780 simulated head impacts, we performed unsupervised domain adaptation on on-field head impacts from 302 college football (CF) impacts and 457 mixed martial arts (MMA) impacts using domain regularized component analysis (DRCA) and cycle-GAN-based methods. The new model improved the MPS/MPSR estimation accuracy, with the DRCA method significantly outperforming other domain adaptation methods in prediction accuracy (p<0.001): MPS RMSE: 0.027 (CF) and 0.037 (MMA); MPSR RMSE: 7.159 (CF) and 13.022 (MMA). On another two hold-out test sets with 195 college football impacts and 260 boxing impacts, the DRCA model significantly outperformed the baseline model without domain adaptation in MPS and MPSR estimation accuracy (p<0.001). The DRCA domain adaptation reduces the MPS/MPSR estimation error to be well below TBI thresholds, enabling accurate brain deformation estimation to detect TBI in future clinical applications.

Abstract: The Variational Autoencoder (VAE) is a seminal approach in deep generative modeling with latent variables. Interpreting its reconstruction process as a nonlinear transformation of samples from the latent posterior distribution, we apply the Unscented Transform (UT) -- a well-known distribution approximation used in the Unscented Kalman Filter (UKF) from the field of filtering. A finite set of statistics called sigma points, sampled deterministically, provides a more informative and lower-variance posterior representation than the ubiquitous noise-scaling of the reparameterization trick, while ensuring higher-quality reconstruction. We further boost the performance by replacing the Kullback-Leibler (KL) divergence with the Wasserstein distribution metric that allows for a sharper posterior. Inspired by the two components, we derive a novel, deterministic-sampling flavor of the VAE, the Unscented Autoencoder (UAE), trained purely with regularization-like terms on the per-sample posterior. We empirically show competitive performance in Fr\'echet Inception Distance (FID) scores over closely-related models, in addition to a lower training variance than the VAE.

Abstract: Recent advances and achievements of artificial intelligence (AI) as well as deep and graph learning models have established their usefulness in biomedical applications, especially in drug-drug interactions (DDIs). DDIs refer to a change in the effect of one drug to the presence of another drug in the human body, which plays an essential role in drug discovery and clinical research. DDIs prediction through traditional clinical trials and experiments is an expensive and time-consuming process. To correctly apply the advanced AI and deep learning, the developer and user meet various challenges such as the availability and encoding of data resources, and the design of computational methods. This review summarizes chemical structure based, network based, NLP based and hybrid methods, providing an updated and accessible guide to the broad researchers and development community with different domain knowledge. We introduce widely-used molecular representation and describe the theoretical frameworks of graph neural network models for representing molecular structures. We present the advantages and disadvantages of deep and graph learning methods by performing comparative experiments. We discuss the potential technical challenges and highlight future directions of deep and graph learning models for accelerating DDIs prediction.

Abstract: In a wide range of multimodal tasks, contrastive learning has become a particularly appealing approach since it can successfully learn representations from abundant unlabeled data with only pairing information (e.g., image-caption or video-audio pairs). Underpinning these approaches is the assumption of multi-view redundancy - that shared information between modalities is necessary and sufficient for downstream tasks. However, in many real-world settings, task-relevant information is also contained in modality-unique regions: information that is only present in one modality but still relevant to the task. How can we learn self-supervised multimodal representations to capture both shared and unique information relevant to downstream tasks? This paper proposes FactorCL, a new multimodal representation learning method to go beyond multi-view redundancy. FactorCL is built from three new contributions: (1) factorizing task-relevant information into shared and unique representations, (2) capturing task-relevant information via maximizing MI lower bounds and removing task-irrelevant information via minimizing MI upper bounds, and (3) multimodal data augmentations to approximate task relevance without labels. On large-scale real-world datasets, FactorCL captures both shared and unique information and achieves state-of-the-art results on six benchmarks.

Abstract: This paper studies federated linear contextual bandits under the notion of user-level differential privacy (DP). We first introduce a unified federated bandits framework that can accommodate various definitions of DP in the sequential decision-making setting. We then formally introduce user-level central DP (CDP) and local DP (LDP) in the federated bandits framework, and investigate the fundamental trade-offs between the learning regrets and the corresponding DP guarantees in a federated linear contextual bandits model. For CDP, we propose a federated algorithm termed as \robin and show that it is near-optimal in terms of the number of clients $M$ and the privacy budget $\varepsilon$ by deriving nearly-matching upper and lower regret bounds when user-level DP is satisfied. For LDP, we obtain several lower bounds, indicating that learning under user-level $(\varepsilon,\delta)$-LDP must suffer a regret blow-up factor at least {$\min\{1/\varepsilon,M\}$ or $\min\{1/\sqrt{\varepsilon},\sqrt{M}\}$} under different conditions.

Abstract: Stochastic gradient descent (SGD) has become a cornerstone of neural network optimization, yet the noise introduced by SGD is often assumed to be uncorrelated over time, despite the ubiquity of epoch-based training. In this work, we challenge this assumption and investigate the effects of epoch-based noise correlations on the stationary distribution of discrete-time SGD with momentum, limited to a quadratic loss. Our main contributions are twofold: first, we calculate the exact autocorrelation of the noise for training in epochs under the assumption that the noise is independent of small fluctuations in the weight vector; second, we explore the influence of correlations introduced by the epoch-based learning scheme on SGD dynamics. We find that for directions with a curvature greater than a hyperparameter-dependent crossover value, the results for uncorrelated noise are recovered. However, for relatively flat directions, the weight variance is significantly reduced. We provide an intuitive explanation for these results based on a crossover between correlation times, contributing to a deeper understanding of the dynamics of SGD in the presence of epoch-based noise correlations.

Abstract: The increasing availability of graph-structured data motivates the task of optimising over functions defined on the node set of graphs. Traditional graph search algorithms can be applied in this case, but they may be sample-inefficient and do not make use of information about the function values; on the other hand, Bayesian optimisation is a class of promising black-box solvers with superior sample efficiency, but it has been scarcely been applied to such novel setups. To fill this gap, we propose a novel Bayesian optimisation framework that optimises over functions defined on generic, large-scale and potentially unknown graphs. Through the learning of suitable kernels on graphs, our framework has the advantage of adapting to the behaviour of the target function. The local modelling approach further guarantees the efficiency of our method. Extensive experiments on both synthetic and real-world graphs demonstrate the effectiveness of the proposed optimisation framework.

Abstract: While Deep Reinforcement Learning has been widely researched in medical imaging, the training and deployment of these models usually require powerful GPUs. Since imaging environments evolve rapidly and can be generated by edge devices, the algorithm is required to continually learn and adapt to changing environments, and adjust to low-compute devices. To this end, we developed three image coreset algorithms to compress and denoise medical images for selective experience replayed-based lifelong reinforcement learning. We implemented neighborhood averaging coreset, neighborhood sensitivity-based sampling coreset, and maximum entropy coreset on full-body DIXON water and DIXON fat MRI images. All three coresets produced 27x compression with excellent performance in localizing five anatomical landmarks: left knee, right trochanter, left kidney, spleen, and lung across both imaging environments. Maximum entropy coreset obtained the best performance of $11.97\pm 12.02$ average distance error, compared to the conventional lifelong learning framework's $19.24\pm 50.77$.

Abstract: This paper addresses intra-client and inter-client covariate shifts in federated learning (FL) with a focus on the overall generalization performance. To handle covariate shifts, we formulate a new global model training paradigm and propose Federated Importance-Weighted Empirical Risk Minimization (FTW-ERM) along with improving density ratio matching methods without requiring perfect knowledge of the supremum over true ratios. We also propose the communication-efficient variant FITW-ERM with the same level of privacy guarantees as those of classical ERM in FL. We theoretically show that FTW-ERM achieves smaller generalization error than classical ERM under certain settings. Experimental results demonstrate the superiority of FTW-ERM over existing FL baselines in challenging imbalanced federated settings in terms of data distribution shifts across clients.

Abstract: Crystal property prediction is a crucial aspect of developing novel materials. However, there are two technical challenges to be addressed for speeding up the investigation of crystals. First, labeling crystal properties is intrinsically difficult due to the high cost and time involved in physical simulations or lab experiments. Second, crystals adhere to a specific quantum chemical principle known as periodic invariance, which is often not captured by existing machine learning methods. To overcome these challenges, we propose the crystal-specific pre-training framework for learning crystal representations with self-supervision. The framework designs a mutex mask strategy for enhancing representation learning so as to alleviate the limited labels available for crystal property prediction. Moreover, we take into account the specific periodic invariance in crystal structures by developing a periodic invariance multi-graph module and periodic attribute learning within our framework. This framework has been tested on eight different tasks. The experimental results on these tasks show that the framework achieves promising prediction performance and is able to outperform recent strong baselines.

Abstract: Test-Time Adaptation (TTA) allows to update pretrained models to changing data distributions at deployment time. While early work tested these algorithms for individual fixed distribution shifts, recent work proposed and applied methods for continual adaptation over long timescales. To examine the reported progress in the field, we propose the Continuously Changing Corruptions (CCC) benchmark to measure asymptotic performance of TTA techniques. We find that eventually all but one state-of-the-art methods collapse and perform worse than a non-adapting model, including models specifically proposed to be robust to performance collapse. In addition, we introduce a simple baseline, "RDumb", that periodically resets the model to its pretrained state. RDumb performs better or on par with the previously proposed state-of-the-art in all considered benchmarks. Our results show that previous TTA approaches are neither effective at regularizing adaptation to avoid collapse nor able to outperform a simplistic resetting strategy.

Abstract: Offline reinforcement learning (RL) is challenged by the distributional shift problem. To address this problem, existing works mainly focus on designing sophisticated policy constraints between the learned policy and the behavior policy. However, these constraints are applied equally to well-performing and inferior actions through uniform sampling, which might negatively affect the learned policy. To alleviate this issue, we propose Offline Prioritized Experience Replay (OPER), featuring a class of priority functions designed to prioritize highly-rewarding transitions, making them more frequently visited during training. Through theoretical analysis, we show that this class of priority functions induce an improved behavior policy, and when constrained to this improved policy, a policy-constrained offline RL algorithm is likely to yield a better solution. We develop two practical strategies to obtain priority weights by estimating advantages based on a fitted value network (OPER-A) or utilizing trajectory returns (OPER-R) for quick computation. OPER is a plug-and-play component for offline RL algorithms. As case studies, we evaluate OPER on five different algorithms, including BC, TD3+BC, Onestep RL, CQL, and IQL. Extensive experiments demonstrate that both OPER-A and OPER-R significantly improve the performance for all baseline methods. Codes and priority weights are availiable at https://github.com/sail-sg/OPER.

Abstract: In this work, we deepen on the use of normalizing flows for causal reasoning. Specifically, we first leverage recent results on non-linear ICA to show that causal models are identifiable from observational data given a causal ordering, and thus can be recovered using autoregressive normalizing flows (NFs). Second, we analyze different design and learning choices for causal normalizing flows to capture the underlying causal data-generating process. Third, we describe how to implement the do-operator in causal NFs, and thus, how to answer interventional and counterfactual questions. Finally, in our experiments, we validate our design and training choices through a comprehensive ablation study; compare causal NFs to other approaches for approximating causal models; and empirically demonstrate that causal NFs can be used to address real-world problems, where the presence of mixed discrete-continuous data and partial knowledge on the causal graph is the norm. The code for this work can be found at https://github.com/psanch21/causal-flows.

Abstract: Spherical CNNs generalize CNNs to functions on the sphere, by using spherical convolutions as the main linear operation. The most accurate and efficient way to compute spherical convolutions is in the spectral domain (via the convolution theorem), which is still costlier than the usual planar convolutions. For this reason, applications of spherical CNNs have so far been limited to small problems that can be approached with low model capacity. In this work, we show how spherical CNNs can be scaled for much larger problems. To achieve this, we make critical improvements including novel variants of common model components, an implementation of core operations to exploit hardware accelerator characteristics, and application-specific input representations that exploit the properties of our model. Experiments show our larger spherical CNNs reach state-of-the-art on several targets of the QM9 molecular benchmark, which was previously dominated by equivariant graph neural networks, and achieve competitive performance on multiple weather forecasting tasks. Our code is available at https://github.com/google-research/spherical-cnn.

Abstract: In many domains, autoregressive models can achieve low log-likelihood on the task of predicting the next observation. However, this maximum-likelihood (MLE) objective does not necessarily match a downstream use-case of autoregressively generating high-quality sequences. The MLE objective weights sequences proportionally to their frequency under the data distribution, with no guidance for the model's behaviour out of distribution (OOD): leading to compounding error during autoregressive generation. In order to address this compounding error problem, we formulate sequence generation as an imitation learning (IL) problem. This allows us to minimize a variety of divergences between the distribution of sequences generated by an autoregressive model and sequences from a dataset, including divergences with weight on OOD generated sequences. The IL framework also allows us to incorporate backtracking by introducing a backspace action into the generation process. This further mitigates the compounding error problem by allowing the model to revert a sampled token if it takes the sequence OOD. Our resulting method, SequenceMatch, can be implemented without adversarial training or major architectural changes. We identify the SequenceMatch-$\chi^2$ divergence as a more suitable training objective for autoregressive models which are used for generation. We show that empirically, SequenceMatch training leads to improvements over MLE on text generation with language models.