Machine Learning (cs.LG)

Abstract: The phenomena of Spectral Bias, where the higher frequency components of a function being learnt in a feedforward Artificial Neural Network (ANN) are seen to converge more slowly than the lower frequencies, is observed ubiquitously across ANNs. This has created technology challenges in fields where resolution of higher frequencies is crucial, like in Physics Informed Neural Networks (PINNs). Extreme Learning Machines (ELMs) that obviate an iterative solution process which provides the theoretical basis of Spectral Bias (SB), should in principle be free of the same. This work verifies the reliability of this assumption, and shows that it is incorrect. However, the structure of ELMs makes them naturally amenable to implementation of variants of Fourier Feature Embeddings, which have been shown to mitigate SB in ANNs. This approach is implemented and verified to completely eliminate SB, thus bringing into feasibility the application of ELMs for practical problems like PINNs where resolution of higher frequencies is essential.

Abstract: We propose a novel generative model for multivariate discrete-time time series data. Drawing inspiration from the construction of neural spline flows, our algorithm incorporates linear transformations and the signature transform as a seamless substitution for traditional neural networks. This approach enables us to achieve not only the universality property inherent in neural networks but also introduces convexity in the model's parameters.

Abstract: Graph neural network (GNN) has been demonstrated powerful in modeling graph-structured data. However, despite many successful cases of applying GNNs to various graph classification and prediction tasks, whether the graph geometrical information has been fully exploited to enhance the learning performance of GNNs is not yet well understood. This paper introduces a new approach to enhance GNN by discrete graph Ricci curvature. Specifically, the graph Ricci curvature defined on the edges of a graph measures how difficult the information transits on one edge from one node to another based on their neighborhoods. Motivated by the geometric analogy of Ricci curvature in the graph setting, we prove that by inserting the curvature information with different carefully designed transformation function $\zeta$, several known computational issues in GNN such as over-smoothing can be alleviated in our proposed model. Furthermore, we verified that edges with very positive Ricci curvature (i.e., $\kappa_{i,j} \approx 1$) are preferred to be dropped to enhance model's adaption to heterophily graph and one curvature based graph edge drop algorithm is proposed. Comprehensive experiments show that our curvature-based GNN model outperforms the state-of-the-art baselines in both homophily and heterophily graph datasets, indicating the effectiveness of involving graph geometric information in GNNs.

Abstract: Which set of features was responsible for a certain output of a machine learning model? Which components caused the failure of a cloud computing application? These are just two examples of questions we are addressing in this work by Identifying Coalition-based Explanations for Common and Rare Events in Any Model (ICECREAM). Specifically, we propose an information-theoretic quantitative measure for the influence of a coalition of variables on the distribution of a target variable. This allows us to identify which set of factors is essential to obtain a certain outcome, as opposed to well-established explainability and causal contribution analysis methods which can assign contributions only to individual factors and rank them by their importance. In experiments with synthetic and real-world data, we show that ICECREAM outperforms state-of-the-art methods for explainability and root cause analysis, and achieves impressive accuracy in both tasks.

Abstract: In the complex domain of large language models (LLMs), striking a balance between computational efficiency and maintaining model quality is a formidable challenge. Navigating the inherent limitations of uniform quantization, particularly when dealing with outliers, and motivated by the launch of NVIDIA's H100 hardware, this study delves into the viability of floating-point (FP) quantization, particularly focusing on FP8 and FP4, as a potential solution. Our comprehensive investigation reveals that for LLMs, FP8 activation consistently outshines its integer (INT8) equivalent, with the performance edge becoming more noticeable in models possessing parameters beyond one billion. For weight quantization, our findings indicate that FP4 exhibits comparable, if not superior, performance to INT4, simplifying deployment on FP-supported hardware like H100. To mitigate the overhead from precision alignment caused by the disparity between weights and activations, we propose two scaling constraints for weight quantization that negligibly impact the performance compared to the standard W4A8 model. We additionally enhance our quantization methods by integrating the Low Rank Compensation (LoRC) strategy, yielding improvements especially in smaller models. The results of our investigation emphasize the immense potential of FP quantization for LLMs, paving the way for high-efficiency deployment in resource-limited settings.

Abstract: In the early observation period of a time series, there might be only a few historic observations available to learn a model. However, in cases where an existing prior set of datasets is available, Meta learning methods can be applicable. In this paper, we devise a Meta learning method that exploits samples from additional datasets and learns to augment time series through adversarial learning as an auxiliary task for the target dataset. Our model (FEML), is equipped with a shared Convolutional backbone that learns features for varying length inputs from different datasets and has dataset specific heads to forecast for different output lengths. We show that FEML can meta learn across datasets and by additionally learning on adversarial generated samples as auxiliary samples for the target dataset, it can improve the forecasting performance compared to single task learning, and various solutions adapted from Joint learning, Multi-task learning and classic forecasting baselines.

Abstract: Obtaining accurate probabilistic forecasts while respecting hierarchical information is an important operational challenge in many applications, perhaps most obviously in energy management, supply chain planning, and resource allocation. The basic challenge, especially for multivariate forecasting, is that forecasts are often required to be coherent with respect to the hierarchical structure. In this paper, we propose a new model which leverages a factor model structure to produce coherent forecasts by construction. This is a consequence of a simple (exchangeability) observation: permuting \textit{}base-level series in the hierarchy does not change their aggregates. Our model uses a convolutional neural network to produce parameters for the factors, their loadings and base-level distributions; it produces samples which can be differentiated with respect to the model's parameters; and it can therefore optimize for any sample-based loss function, including the Continuous Ranked Probability Score and quantile losses. We can choose arbitrary continuous distributions for the factor and the base-level distributions. We compare our method to two previous methods which can be optimized end-to-end, while enforcing coherent aggregation. Our model achieves significant improvements: between $11.8-41.4\%$ on three hierarchical forecasting datasets. We also analyze the influence of parameters in our model with respect to base-level distribution and number of factors.

Abstract: Graph learning has a wide range of applications in many scenarios, which require more need for data privacy. Federated learning is an emerging distributed machine learning approach that leverages data from individual devices or data centers to improve the accuracy and generalization of the model, while also protecting the privacy of user data. Graph-federated learning is mainly based on the classical federated learning framework i.e., the Client-Server framework. However, the Client-Server framework faces problems such as a single point of failure of the central server and poor scalability of network topology. First, we introduce the decentralized framework to graph-federated learning. Second, determine the confidence among nodes based on the similarity of data among nodes, subsequently, the gradient information is then aggregated by linear weighting based on confidence. Finally, the proposed method is compared with FedAvg, Fedprox, GCFL, and GCFL+ to verify the effectiveness of the proposed method. Experiments demonstrate that the proposed method outperforms other methods.

Abstract: Frequency analysis is useful for understanding the mechanisms of representation learning in neural networks (NNs). Most research in this area focuses on the learning dynamics of NNs for regression tasks, while little for classification. This study empirically investigates the latter and expands the understanding of frequency shortcuts. First, we perform experiments on synthetic datasets, designed to have a bias in different frequency bands. Our results demonstrate that NNs tend to find simple solutions for classification, and what they learn first during training depends on the most distinctive frequency characteristics, which can be either low- or high-frequencies. Second, we confirm this phenomenon on natural images. We propose a metric to measure class-wise frequency characteristics and a method to identify frequency shortcuts. The results show that frequency shortcuts can be texture-based or shape-based, depending on what best simplifies the objective. Third, we validate the transferability of frequency shortcuts on out-of-distribution (OOD) test sets. Our results suggest that frequency shortcuts can be transferred across datasets and cannot be fully avoided by larger model capacity and data augmentation. We recommend that future research should focus on effective training schemes mitigating frequency shortcut learning.

Abstract: Looking for sparsity is nowadays crucial to speed up the training of large-scale neural networks. Projections onto the $\ell_{1,2}$ and $\ell_{1,\infty}$ are among the most efficient techniques to sparsify and reduce the overall cost of neural networks. In this paper, we introduce a new projection algorithm for the $\ell_{1,\infty}$ norm ball. The worst-case time complexity of this algorithm is $\mathcal{O}\big(nm+J\log(nm)\big)$ for a matrix in $\mathbb{R}^{n\times m}$. $J$ is a term that tends to 0 when the sparsity is high, and to $nm$ when the sparsity is low. Its implementation is easy and it is guaranteed to converge to the exact solution in a finite time. Moreover, we propose to incorporate the $\ell_{1,\infty}$ ball projection while training an autoencoder to enforce feature selection and sparsity of the weights. Sparsification appears in the encoder to primarily do feature selection due to our application in biology, where only a very small part ($<2\%$) of the data is relevant. We show that both in the biological case and in the general case of sparsity that our method is the fastest.

Abstract: Machine learning has affected the way in which many phenomena for various domains are modelled, one of these domains being that of structural dynamics. However, because machine-learning algorithms are problem-specific, they often fail to perform efficiently in cases of data scarcity. To deal with such issues, combination of physics-based approaches and machine learning algorithms have been developed. Although such methods are effective, they also require the analyser's understanding of the underlying physics of the problem. The current work is aimed at motivating the use of models which learn such relationships from a population of phenomena, whose underlying physics are similar. The development of such models is motivated by the way that physics-based models, and more specifically finite element models, work. Such models are considered transferrable, explainable and trustworthy, attributes which are not trivially imposed or achieved for machine-learning models. For this reason, machine-learning approaches are less trusted by industry and often considered more difficult to form validated models. To achieve such data-driven models, a population-based scheme is followed here and two different machine-learning algorithms from the meta-learning domain are used. The two algorithms are the model-agnostic meta-learning (MAML) algorithm and the conditional neural processes (CNP) model. The algorithms seem to perform as intended and outperform a traditional machine-learning algorithm at approximating the quantities of interest. Moreover, they exhibit behaviour similar to traditional machine learning algorithms (e.g. neural networks or Gaussian processes), concerning their performance as a function of the available structures in the training population.

Abstract: Understanding and characterizing the vulnerability of urban infrastructures, which refers to the engineering facilities essential for the regular running of cities and that exist naturally in the form of networks, is of great value to us. Potential applications include protecting fragile facilities and designing robust topologies, etc. Due to the strong correlation between different topological characteristics and infrastructure vulnerability and their complicated evolution mechanisms, some heuristic and machine-assisted analysis fall short in addressing such a scenario. In this paper, we model the interdependent network as a heterogeneous graph and propose a system based on graph neural network with reinforcement learning, which can be trained on real-world data, to characterize the vulnerability of the city system accurately. The presented system leverages deep learning techniques to understand and analyze the heterogeneous graph, which enables us to capture the risk of cascade failure and discover vulnerable infrastructures of cities. Extensive experiments with various requests demonstrate not only the expressive power of our system but also transferring ability and necessity of the specific components.

Abstract: Generative Adversarial Networks (GANs) can produce high-quality samples, but do not provide an estimate of the probability density around the samples. However, it has been noted that maximizing the log-likelihood within an energy-based setting can lead to an adversarial framework where the discriminator provides unnormalized density (often called energy). We further develop this perspective, incorporate importance sampling, and show that 1) Wasserstein GAN performs a biased estimate of the partition function, and we propose instead to use an unbiased estimator; 2) when optimizing for likelihood, one must maximize generator entropy. This is hypothesized to provide a better mode coverage. Different from previous works, we explicitly compute the density of the generated samples. This is the key enabler to designing an unbiased estimator of the partition function and computation of the generator entropy term. The generator density is obtained via a new type of flow network, called one-way flow network, that is less constrained in terms of architecture, as it does not require to have a tractable inverse function. Our experimental results show that we converge faster, produce comparable sample quality to GANs with similar architecture, successfully avoid over-fitting to commonly used datasets and produce smooth low-dimensional latent representations of the training data.

Abstract: We view variational autoencoders (VAE) as decoder-encoder pairs, which map distributions in the data space to distributions in the latent space and vice versa. The standard learning approach for VAEs, i.e. maximisation of the evidence lower bound (ELBO), has an obvious asymmetry in that respect. Moreover, it requires a closed form a-priori latent distribution. This limits the applicability of VAEs in more complex scenarios, such as general semi-supervised learning and employing complex generative models as priors. We propose a Nash equilibrium learning approach that relaxes these restrictions and allows learning VAEs in situations where both the data and the latent distributions are accessible only by sampling. The flexibility and simplicity of this approach allows its application to a wide range of learning scenarios and downstream tasks. We show experimentally that the models learned by this method are comparable to those obtained by ELBO learning and demonstrate its applicability for tasks that are not accessible by standard VAE learning.

Abstract: We consider the general class of time-homogeneous dynamical systems, both discrete and continuous, and study the problem of learning a meaningful representation of the state from observed data. This is instrumental for the task of learning a forward transfer operator of the system, that in turn can be used for forecasting future states or observables. The representation, typically parametrized via a neural network, is associated with a projection operator and is learned by optimizing an objective function akin to that of canonical correlation analysis (CCA). However, unlike CCA, our objective avoids matrix inversions and therefore is generally more stable and applicable to challenging scenarios. Our objective is a tight relaxation of CCA and we further enhance it by proposing two regularization schemes, one encouraging the orthogonality of the components of the representation while the other exploiting Chapman-Kolmogorov's equation. We apply our method to challenging discrete dynamical systems, discussing improvements over previous methods, as well as to continuous dynamical systems.

Abstract: To avoid failures on out-of-distribution data, recent works have sought to extract features that have a stable or invariant relationship with the label across domains, discarding the "spurious" or unstable features whose relationship with the label changes across domains. However, unstable features often carry complementary information about the label that could boost performance if used correctly in the test domain. Our main contribution is to show that it is possible to learn how to use these unstable features in the test domain without labels. In particular, we prove that pseudo-labels based on stable features provide sufficient guidance for doing so, provided that stable and unstable features are conditionally independent given the label. Based on this theoretical insight, we propose Stable Feature Boosting (SFB), an algorithm for: (i) learning a predictor that separates stable and conditionally-independent unstable features; and (ii) using the stable-feature predictions to adapt the unstable-feature predictions in the test domain. Theoretically, we prove that SFB can learn an asymptotically-optimal predictor without test-domain labels. Empirically, we demonstrate the effectiveness of SFB on real and synthetic data.

Abstract: Clinical trials are critical for drug development but often suffer from expensive and inefficient patient recruitment. In recent years, machine learning models have been proposed for speeding up patient recruitment via automatically matching patients with clinical trials based on longitudinal patient electronic health records (EHR) data and eligibility criteria of clinical trials. However, they either depend on trial-specific expert rules that cannot expand to other trials or perform matching at a very general level with a black-box model where the lack of interpretability makes the model results difficult to be adopted. To provide accurate and interpretable patient trial matching, we introduce a personalized dynamic tree-based memory network model named TREEMENT. It utilizes hierarchical clinical ontologies to expand the personalized patient representation learned from sequential EHR data, and then uses an attentional beam-search query learned from eligibility criteria embedding to offer a granular level of alignment for improved performance and interpretability. We evaluated TREEMENT against existing models on real-world datasets and demonstrated that TREEMENT outperforms the best baseline by 7% in terms of error reduction in criteria-level matching and achieves state-of-the-art results in its trial-level matching ability. Furthermore, we also show TREEMENT can offer good interpretability to make the model results easier for adoption.

Abstract: Recommender systems are a ubiquitous feature of online platforms. Increasingly, they are explicitly tasked with increasing users' long-term satisfaction. In this context, we study a content exploration task, which we formalize as a multi-armed bandit problem with delayed rewards. We observe that there is an apparent trade-off in choosing the learning signal: Waiting for the full reward to become available might take several weeks, hurting the rate at which learning happens, whereas measuring short-term proxy rewards reflects the actual long-term goal only imperfectly. We address this challenge in two steps. First, we develop a predictive model of delayed rewards that incorporates all information obtained to date. Full observations as well as partial (short or medium-term) outcomes are combined through a Bayesian filter to obtain a probabilistic belief. Second, we devise a bandit algorithm that takes advantage of this new predictive model. The algorithm quickly learns to identify content aligned with long-term success by carefully balancing exploration and exploitation. We apply our approach to a podcast recommendation problem, where we seek to identify shows that users engage with repeatedly over two months. We empirically validate that our approach results in substantially better performance compared to approaches that either optimize for short-term proxies, or wait for the long-term outcome to be fully realized.

Abstract: The paradigm of federated learning (FL) to address data privacy concerns by locally training parameters on resource-constrained clients in a distributed manner has garnered significant attention. Nonetheless, FL is not applicable when not all clients within the coverage of the FL server are registered with the FL network. To bridge this gap, this paper proposes joint learner referral aided federated client selection (LRef-FedCS), along with communications and computing resource scheduling, and local model accuracy optimization (LMAO) methods. These methods are designed to minimize the cost incurred by the worst-case participant and ensure the long-term fairness of FL in hierarchical Internet of Things (HieIoT) networks. Utilizing the Lyapunov optimization technique, we reformulate the original problem into a stepwise joint optimization problem (JOP). Subsequently, to tackle the mixed-integer non-convex JOP, we separatively and iteratively address LRef-FedCS and LMAO through the centralized method and self-adaptive global best harmony search (SGHS) algorithm, respectively. To enhance scalability, we further propose a distributed LRef-FedCS approach based on a matching game to replace the centralized method described above. Numerical simulations and experimental results on the MNIST/CIFAR-10 datasets demonstrate that our proposed LRef-FedCS approach could achieve a good balance between pursuing high global accuracy and reducing cost.

Abstract: On-device training is essential for user personalisation and privacy. With the pervasiveness of IoT devices and microcontroller units (MCU), this task becomes more challenging due to the constrained memory and compute resources, and the limited availability of labelled user data. Nonetheless, prior works neglect the data scarcity issue, require excessively long training time (e.g. a few hours), or induce substantial accuracy loss ($\geq$10\%). We propose TinyTrain, an on-device training approach that drastically reduces training time by selectively updating parts of the model and explicitly coping with data scarcity. TinyTrain introduces a task-adaptive sparse-update method that dynamically selects the layer/channel based on a multi-objective criterion that jointly captures user data, the memory, and the compute capabilities of the target device, leading to high accuracy on unseen tasks with reduced computation and memory footprint. TinyTrain outperforms vanilla fine-tuning of the entire network by 3.6-5.0\% in accuracy, while reducing the backward-pass memory and computation cost by up to 2,286$\times$ and 7.68$\times$, respectively. Targeting broadly used real-world edge devices, TinyTrain achieves 9.5$\times$ faster and 3.5$\times$ more energy-efficient training over status-quo approaches, and 2.8$\times$ smaller memory footprint than SOTA approaches, while remaining within the 1 MB memory envelope of MCU-grade platforms.

Abstract: Deploying deep learning neural networks on edge devices, to accomplish task specific objectives in the real-world, requires a reduction in their memory footprint, power consumption, and latency. This can be realized via efficient model compression. Disentangled latent representations produced by variational autoencoder (VAE) networks are a promising approach for achieving model compression because they mainly retain task-specific information, discarding useless information for the task at hand. We make use of the Beta-VAE framework combined with a standard criterion for pruning to investigate the impact of forcing the network to learn disentangled representations on the pruning process for the task of classification. In particular, we perform experiments on MNIST and CIFAR10 datasets, examine disentanglement challenges, and propose a path forward for future works.

Abstract: I propose an open dataset of country-level historical opinion polling data for the European Union and the UK. The dataset aims to fill a gap in available opinion polling data for the European Union. Some existing datasets are restricted to the past five years, limiting research opportunities. At the same time, some larger proprietary datasets exist but are available only in a visual preprocessed time series format. Finally, while other large datasets for individual countries might exist, these could be inaccessible due to language barriers. The data was gathered from Wikipedia, and preprocessed using the pandas library. Both the raw and the preprocessed data are in the .csv format. I hope that given the recent advances in LLMs and deep learning in general, this large dataset will enable researchers to uncover complex interactions between multimodal data (news articles, economic indicators, social media) and voting behavior. The raw data, the preprocessed data, and the preprocessing scripts are available on GitHub.

Abstract: Deep neural networks often fail catastrophically by relying on spurious correlations. Most prior work assumes a clear dichotomy into spurious and reliable features; however, this is often unrealistic. For example, most of the time we do not want an autonomous car to simply copy the speed of surrounding cars -- we don't want our car to run a red light if a neighboring car does so. However, we cannot simply enforce invariance to next-lane speed, since it could provide valuable information about an unobservable pedestrian at a crosswalk. Thus, universally ignoring features that are sometimes (but not always) reliable can lead to non-robust performance. We formalize a new setting called contextual reliability which accounts for the fact that the "right" features to use may vary depending on the context. We propose and analyze a two-stage framework called Explicit Non-spurious feature Prediction (ENP) which first identifies the relevant features to use for a given context, then trains a model to rely exclusively on these features. Our work theoretically and empirically demonstrates the advantages of ENP over existing methods and provides new benchmarks for contextual reliability.

Abstract: In this paper, we characterize Probabilistic Principal Component Analysis in Hilbert spaces and demonstrate how the optimal solution admits a representation in dual space. This allows us to develop a generative framework for kernel methods. Furthermore, we show how it englobes Kernel Principal Component Analysis and illustrate its working on a toy and a real dataset.

Abstract: There is a growing interest in device-control systems that can interpret human natural language instructions and execute them on a digital device by directly controlling its user interface. We present a dataset for device-control research, Android in the Wild (AITW), which is orders of magnitude larger than current datasets. The dataset contains human demonstrations of device interactions, including the screens and actions, and corresponding natural language instructions. It consists of 715k episodes spanning 30k unique instructions, four versions of Android (v10-13),and eight device types (Pixel 2 XL to Pixel 6) with varying screen resolutions. It contains multi-step tasks that require semantic understanding of language and visual context. This dataset poses a new challenge: actions available through the user interface must be inferred from their visual appearance. And, instead of simple UI element-based actions, the action space consists of precise gestures (e.g., horizontal scrolls to operate carousel widgets). We organize our dataset to encourage robustness analysis of device-control systems, i.e., how well a system performs in the presence of new task descriptions, new applications, or new platform versions. We develop two agents and report performance across the dataset. The dataset is available at https://github.com/google-research/google-research/tree/master/android_in_the_wild.

Abstract: Gromov-Wasserstein distance has found many applications in machine learning due to its ability to compare measures across metric spaces and its invariance to isometric transformations. However, in certain applications, this invariance property can be too flexible, thus undesirable. Moreover, the Gromov-Wasserstein distance solely considers pairwise sample similarities in input datasets, disregarding the raw feature representations. We propose a new optimal transport-based distance, called Augmented Gromov-Wasserstein, that allows for some control over the level of rigidity to transformations. It also incorporates feature alignments, enabling us to better leverage prior knowledge on the input data for improved performance. We present theoretical insights into the proposed metric. We then demonstrate its usefulness for single-cell multi-omic alignment tasks and a transfer learning scenario in machine learning.

Abstract: Movement paths are used widely in intelligent transportation and smart city applications. To serve such applications, path representation learning aims to provide compact representations of paths that enable efficient and accurate operations when used for different downstream tasks such as path ranking and travel cost estimation. In many cases, it is attractive that the path representation learning is lightweight and scalable; in resource-limited environments and under green computing limitations, it is essential. Yet, existing path representation learning studies focus on accuracy and pay at most secondary attention to resource consumption and scalability. We propose a lightweight and scalable path representation learning framework, termed LightPath, that aims to reduce resource consumption and achieve scalability without affecting accuracy, thus enabling broader applicability. More specifically, we first propose a sparse auto-encoder that ensures that the framework achieves good scalability with respect to path length. Next, we propose a relational reasoning framework to enable faster training of more robust sparse path encoders. We also propose global-local knowledge distillation to further reduce the size and improve the performance of sparse path encoders. Finally, we report extensive experiments on two real-world datasets to offer insight into the efficiency, scalability, and effectiveness of the proposed framework.