Machine Learning (cs.LG)

Abstract: Electronic Health Record (EHR) data frequently exhibits sparse characteristics, posing challenges for predictive modeling. Current direct imputation such as matrix imputation approaches hinge on referencing analogous rows or columns to complete raw missing data and do not differentiate between imputed and actual values. As a result, models may inadvertently incorporate irrelevant or deceptive information with respect to the prediction objective, thereby compromising the efficacy of downstream performance. While some methods strive to recalibrate or augment EHR embeddings after direct imputation, they often mistakenly prioritize imputed features. This misprioritization can introduce biases or inaccuracies into the model. To tackle these issues, our work resorts to indirect imputation, where we leverage prototype representations from similar patients to obtain a denser embedding. Recognizing the limitation that missing features are typically treated the same as present ones when measuring similar patients, our approach designs a feature confidence learner module. This module is sensitive to the missing feature status, enabling the model to better judge the reliability of each feature. Moreover, we propose a novel patient similarity metric that takes feature confidence into account, ensuring that evaluations are not based merely on potentially inaccurate imputed values. Consequently, our work captures dense prototype patient representations with feature-missing-aware calibration process. Comprehensive experiments demonstrate that designed model surpasses established EHR-focused models with a statistically significant improvement on MIMIC-III and MIMIC-IV datasets in-hospital mortality outcome prediction task. The code is publicly available at \url{https://anonymous.4open.science/r/SparseEHR} to assure the reproducibility.

Abstract: Dataset distillation methods have demonstrated remarkable performance for neural networks trained with very limited training data. However, a significant challenge arises in the form of architecture overfitting: the distilled training data synthesized by a specific network architecture (i.e., training network) generates poor performance when trained by other network architectures (i.e., test networks). This paper addresses this issue and proposes a series of approaches in both architecture designs and training schemes which can be adopted together to boost the generalization performance across different network architectures on the distilled training data. We conduct extensive experiments to demonstrate the effectiveness and generality of our methods. Particularly, across various scenarios involving different sizes of distilled data, our approaches achieve comparable or superior performance to existing methods when training on the distilled data using networks with larger capacities.

Abstract: Counterfactuals operationalised through algorithmic recourse have become a powerful tool to make artificial intelligence systems explainable. Conceptually, given an individual classified as y -- the factual -- we seek actions such that their prediction becomes the desired class y' -- the counterfactual. This process offers algorithmic recourse that is (1) easy to customise and interpret, and (2) directly aligned with the goals of each individual. However, the properties of a "good" counterfactual are still largely debated; it remains an open challenge to effectively locate a counterfactual along with its corresponding recourse. Some strategies use gradient-driven methods, but these offer no guarantees on the feasibility of the recourse and are open to adversarial attacks on carefully created manifolds. This can lead to unfairness and lack of robustness. Other methods are data-driven, which mostly addresses the feasibility problem at the expense of privacy, security and secrecy as they require access to the entire training data set. Here, we introduce LocalFACE, a model-agnostic technique that composes feasible and actionable counterfactual explanations using locally-acquired information at each step of the algorithmic recourse. Our explainer preserves the privacy of users by only leveraging data that it specifically requires to construct actionable algorithmic recourse, and protects the model by offering transparency solely in the regions deemed necessary for the intervention.

Abstract: Off-Policy Estimation (OPE) methods allow us to learn and evaluate decision-making policies from logged data. This makes them an attractive choice for the offline evaluation of recommender systems, and several recent works have reported successful adoption of OPE methods to this end. An important assumption that makes this work is the absence of unobserved confounders: random variables that influence both actions and rewards at data collection time. Because the data collection policy is typically under the practitioner's control, the unconfoundedness assumption is often left implicit, and its violations are rarely dealt with in the existing literature. This work aims to highlight the problems that arise when performing off-policy estimation in the presence of unobserved confounders, specifically focusing on a recommendation use-case. We focus on policy-based estimators, where the logging propensities are learned from logged data. We characterise the statistical bias that arises due to confounding, and show how existing diagnostics are unable to uncover such cases. Because the bias depends directly on the true and unobserved logging propensities, it is non-identifiable. As the unconfoundedness assumption is famously untestable, this becomes especially problematic. This paper emphasises this common, yet often overlooked issue. Through synthetic data, we empirically show how na\"ive propensity estimation under confounding can lead to severely biased metric estimates that are allowed to fly under the radar. We aim to cultivate an awareness among researchers and practitioners of this important problem, and touch upon potential research directions towards mitigating its effects.

Abstract: Data silos, mainly caused by privacy and interoperability, significantly constrain collaborations among different organizations with similar data for the same purpose. Distributed learning based on divide-and-conquer provides a promising way to settle the data silos, but it suffers from several challenges, including autonomy, privacy guarantees, and the necessity of collaborations. This paper focuses on developing an adaptive distributed kernel ridge regression (AdaDKRR) by taking autonomy in parameter selection, privacy in communicating non-sensitive information, and the necessity of collaborations in performance improvement into account. We provide both solid theoretical verification and comprehensive experiments for AdaDKRR to demonstrate its feasibility and effectiveness. Theoretically, we prove that under some mild conditions, AdaDKRR performs similarly to running the optimal learning algorithms on the whole data, verifying the necessity of collaborations and showing that no other distributed learning scheme can essentially beat AdaDKRR under the same conditions. Numerically, we test AdaDKRR on both toy simulations and two real-world applications to show that AdaDKRR is superior to other existing distributed learning schemes. All these results show that AdaDKRR is a feasible scheme to defend against data silos, which are highly desired in numerous application regions such as intelligent decision-making, pricing forecasting, and performance prediction for products.

Abstract: There are now many comprehension algorithms for understanding the decisions of a machine learning algorithm. Among these are those based on the generation of counterfactual examples. This article proposes to view this generation process as a source of creating a certain amount of knowledge that can be stored to be used, later, in different ways. This process is illustrated in the additive model and, more specifically, in the case of the naive Bayes classifier, whose interesting properties for this purpose are shown.

Abstract: In traditional deep learning algorithms, one of the key assumptions is that the data distribution remains constant during both training and deployment. However, this assumption becomes problematic when faced with Out-of-Distribution periods, such as the COVID-19 lockdowns, where the data distribution significantly deviates from what the model has seen during training. This paper employs a two-fold strategy: utilizing continual learning techniques to update models with new data and harnessing human mobility data collected from privacy-preserving pedestrian counters located outside buildings. In contrast to online learning, which suffers from 'catastrophic forgetting' as newly acquired knowledge often erases prior information, continual learning offers a holistic approach by preserving past insights while integrating new data. This research applies FSNet, a powerful continual learning algorithm, to real-world data from 13 building complexes in Melbourne, Australia, a city which had the second longest total lockdown duration globally during the pandemic. Results underscore the crucial role of continual learning in accurate energy forecasting, particularly during Out-of-Distribution periods. Secondary data such as mobility and temperature provided ancillary support to the primary forecasting model. More importantly, while traditional methods struggled to adapt during lockdowns, models featuring at least online learning demonstrated resilience, with lockdown periods posing fewer challenges once armed with adaptive learning techniques. This study contributes valuable methodologies and insights to the ongoing effort to improve energy load forecasting during future Out-of-Distribution periods.

Abstract: The provision of social care applications is crucial for elderly people to improve their quality of life and enables operators to provide early interventions. Accurate predictions of user dropouts in healthy ageing applications are essential since they are directly related to individual health statuses. Machine Learning (ML) algorithms have enabled highly accurate predictions, outperforming traditional statistical methods that struggle to cope with individual patterns. However, ML requires a substantial amount of data for training, which is challenging due to the presence of personal identifiable information (PII) and the fragmentation posed by regulations. In this paper, we present a federated machine learning (FML) approach that minimizes privacy concerns and enables distributed training, without transferring individual data. We employ collaborative training by considering individuals and organizations under FML, which models both cross-device and cross-silo learning scenarios. Our approach is evaluated on a real-world dataset with non-independent and identically distributed (non-iid) data among clients, class imbalance and label ambiguity. Our results show that data selection and class imbalance handling techniques significantly improve the predictive accuracy of models trained under FML, demonstrating comparable or superior predictive performance than traditional ML models.

Abstract: Most real-world classification tasks suffer from label noise to some extent. Such noise in the data adversely affects the generalization error of learned models and complicates the evaluation of noise-handling methods, as their performance cannot be accurately measured without clean labels. In label noise research, typically either noisy or incomplex simulated data are accepted as a baseline, into which additional noise with known properties is injected. In this paper, we propose SYNLABEL, a framework that aims to improve upon the aforementioned methodologies. It allows for creating a noiseless dataset informed by real data, by either pre-specifying or learning a function and defining it as the ground truth function from which labels are generated. Furthermore, by resampling a number of values for selected features in the function domain, evaluating the function and aggregating the resulting labels, each data point can be assigned a soft label or label distribution. Such distributions allow for direct injection and quantification of label noise. The generated datasets serve as a clean baseline of adjustable complexity into which different types of noise may be introduced. We illustrate how the framework can be applied, how it enables quantification of label noise and how it improves over existing methodologies.

Abstract: Predictions over graphs play a crucial role in various domains, including social networks, molecular biology, medicine, and more. Graph Neural Networks (GNNs) have emerged as the dominant approach for learning on graph data. Instances of graph labeling problems consist of the graph-structure (i.e., the adjacency matrix), along with node-specific feature vectors. In some cases, this graph-structure is non-informative for the predictive task. For instance, molecular properties such as molar mass depend solely on the constituent atoms (node features), and not on the molecular structure. While GNNs have the ability to ignore the graph-structure in such cases, it is not clear that they will. In this work, we show that GNNs actually tend to overfit the graph-structure in the sense that they use it even when a better solution can be obtained by ignoring it. We examine this phenomenon with respect to different graph distributions and find that regular graphs are more robust to this overfitting. We then provide a theoretical explanation for this phenomenon, via analyzing the implicit bias of gradient-descent-based learning of GNNs in this setting. Finally, based on our empirical and theoretical findings, we propose a graph-editing method to mitigate the tendency of GNNs to overfit graph-structures that should be ignored. We show that this method indeed improves the accuracy of GNNs across multiple benchmarks.

Abstract: We study monotone submodular maximization under general matroid constraints in the online setting. We prove that online optimization of a large class of submodular functions, namely, weighted threshold potential functions, reduces to online convex optimization (OCO). This is precisely because functions in this class admit a concave relaxation; as a result, OCO policies, coupled with an appropriate rounding scheme, can be used to achieve sublinear regret in the combinatorial setting. We show that our reduction extends to many different versions of the online learning problem, including the dynamic regret, bandit, and optimistic-learning settings.

Abstract: Zero-shot inference is a powerful paradigm that enables the use of large pretrained models for downstream classification tasks without further training. However, these models are vulnerable to inherited biases that can impact their performance. The traditional solution is fine-tuning, but this undermines the key advantage of pretrained models, which is their ability to be used out-of-the-box. We propose RoboShot, a method that improves the robustness of pretrained model embeddings in a fully zero-shot fashion. First, we use zero-shot language models (LMs) to obtain useful insights from task descriptions. These insights are embedded and used to remove harmful and boost useful components in embeddings -- without any supervision. Theoretically, we provide a simple and tractable model for biases in zero-shot embeddings and give a result characterizing under what conditions our approach can boost performance. Empirically, we evaluate RoboShot on nine image and NLP classification tasks and show an average improvement of 15.98% over several zero-shot baselines. Additionally, we demonstrate that RoboShot is compatible with a variety of pretrained and language models.

Abstract: As power quality becomes a higher priority in the electric utility industry, the amount of disturbance event data continues to grow. Utilities do not have the required personnel to analyze each event by hand. This work presents an automated approach for analyzing power quality events recorded by digital fault recorders and power quality monitors operating within a power transmission system. The automated approach leverages rule-based analytics to examine the time and frequency domain characteristics of the voltage and current signals. Customizable thresholds are set to categorize each disturbance event. The events analyzed within this work include various faults, motor starting, and incipient instrument transformer failure. Analytics for fourteen different event types have been developed. The analytics were tested on 160 signal files and yielded an accuracy of ninety-nine percent. Continuous, nominal signal data analysis is performed using an approach coined as the cyclic histogram. The cyclic histogram process will be integrated into the digital fault recorders themselves to facilitate the detection of subtle signal variations that are too small to trigger a disturbance event and that can occur over hours or days. In addition to reducing memory requirements by a factor of 320, it is anticipated that cyclic histogram processing will aid in identifying incipient events and identifiers. This project is expected to save engineers time by automating the classification of disturbance events and increase the reliability of the transmission system by providing near real time detection and identification of disturbances as well as prevention of problems before they occur.

Abstract: Determining the completability of levels generated by procedural generators such as machine learning models can be challenging, as it can involve the use of solver agents that often require a significant amount of time to analyze and solve levels. Active learning is not yet widely adopted in game evaluations, although it has been used successfully in natural language processing, image and speech recognition, and computer vision, where the availability of labeled data is limited or expensive. In this paper, we propose the use of active learning for learning level completability classification. Through an active learning approach, we train deep-learning models to classify the completability of generated levels for Super Mario Bros., Kid Icarus, and a Zelda-like game. We compare active learning for querying levels to label with completability against random queries. Our results show using an active learning approach to label levels results in better classifier performance with the same amount of labeled data.

Abstract: A fundamental question in theoretical machine learning is generalization. Over the past decades, the PAC-Bayesian approach has been established as a flexible framework to address the generalization capabilities of machine learning algorithms, and design new ones. Recently, it has garnered increased interest due to its potential applicability for a variety of learning algorithms, including deep neural networks. In parallel, an information-theoretic view of generalization has developed, wherein the relation between generalization and various information measures has been established. This framework is intimately connected to the PAC-Bayesian approach, and a number of results have been independently discovered in both strands. In this monograph, we highlight this strong connection and present a unified treatment of generalization. We present techniques and results that the two perspectives have in common, and discuss the approaches and interpretations that differ. In particular, we demonstrate how many proofs in the area share a modular structure, through which the underlying ideas can be intuited. We pay special attention to the conditional mutual information (CMI) framework; analytical studies of the information complexity of learning algorithms; and the application of the proposed methods to deep learning. This monograph is intended to provide a comprehensive introduction to information-theoretic generalization bounds and their connection to PAC-Bayes, serving as a foundation from which the most recent developments are accessible. It is aimed broadly towards researchers with an interest in generalization and theoretical machine learning.

Abstract: Several domains increasingly rely on machine learning in their applications. The resulting heavy dependence on data has led to the emergence of various laws and regulations around data ethics and privacy and growing awareness of the need for privacy-preserving machine learning (ppML). Current ppML techniques utilize methods that are either purely based on cryptography, such as homomorphic encryption, or that introduce noise into the input, such as differential privacy. The main criticism given to those techniques is the fact that they either are too slow or they trade off a model s performance for improved confidentiality. To address this performance reduction, we aim to leverage robust representation learning as a way of encoding our data while optimizing the privacy-utility trade-off. Our method centers on training autoencoders in a multi-objective manner and then concatenating the latent and learned features from the encoding part as the encoded form of our data. Such a deep learning-powered encoding can then safely be sent to a third party for intensive training and hyperparameter tuning. With our proposed framework, we can share our data and use third party tools without being under the threat of revealing its original form. We empirically validate our results on unimodal and multimodal settings, the latter following a vertical splitting system and show improved performance over state-of-the-art.

Abstract: Normalizing Flows (NFs) describe a class of models that express a complex target distribution as the composition of a series of bijective transformations over a simpler base distribution. By limiting the space of candidate transformations to diffeomorphisms, NFs enjoy efficient, exact sampling and density evaluation, enabling NFs to flexibly behave as both discriminative and generative models. Their restriction to diffeomorphisms, however, enforces that input, output and all intermediary spaces share the same dimension, limiting their ability to effectively represent target distributions with complex topologies. Additionally, in cases where the prior and target distributions are not homeomorphic, Normalizing Flows can leak mass outside of the support of the target. This survey covers a selection of recent works that combine aspects of other generative model classes, such as VAEs and score-based diffusion, and in doing so loosen the strict bijectivity constraints of NFs to achieve a balance of expressivity, training speed, sample efficiency and likelihood tractability.

Abstract: Exploration in sparse-reward reinforcement learning is difficult due to the requirement of long, coordinated sequences of actions in order to achieve any reward. Moreover, in continuous action spaces there are an infinite number of possible actions, which only increases the difficulty of exploration. One class of methods designed to address these issues forms temporally extended actions, often called skills, from interaction data collected in the same domain, and optimizes a policy on top of this new action space. Typically such methods require a lengthy pretraining phase, especially in continuous action spaces, in order to form the skills before reinforcement learning can begin. Given prior evidence that the full range of the continuous action space is not required in such tasks, we propose a novel approach to skill-generation with two components. First we discretize the action space through clustering, and second we leverage a tokenization technique borrowed from natural language processing to generate temporally extended actions. Such a method outperforms baselines for skill-generation in several challenging sparse-reward domains, and requires orders-of-magnitude less computation in skill-generation and online rollouts.

Abstract: Predicting future outcomes is a prevalent application of machine learning in social impact domains. Examples range from predicting student success in education to predicting disease risk in healthcare. Practitioners recognize that the ultimate goal is not just to predict but to act effectively. Increasing evidence suggests that relying on outcome predictions for downstream interventions may not have desired results. In most domains there exists a multitude of possible interventions for each individual, making the challenge of taking effective action more acute. Even when causal mechanisms connecting the individual's latent states to outcomes is well understood, in any given instance (a specific student or patient), practitioners still need to infer -- from budgeted measurements of latent states -- which of many possible interventions will be most effective for this individual. With this in mind, we ask: when are accurate predictors of outcomes helpful for identifying the most suitable intervention? Through a simple model encompassing actions, latent states, and measurements, we demonstrate that pure outcome prediction rarely results in the most effective policy for taking actions, even when combined with other measurements. We find that except in cases where there is a single decisive action for improving the outcome, outcome prediction never maximizes "action value", the utility of taking actions. Making measurements of actionable latent states, where specific actions lead to desired outcomes, considerably enhances the action value compared to outcome prediction, and the degree of improvement depends on action costs and the outcome model. This analysis emphasizes the need to go beyond generic outcome prediction in interventional settings by incorporating knowledge of plausible actions and latent states.