Tighter Bounds on the Information Bottleneck with Application to Deep Learning

Deep Neural Nets (DNNs) learn latent representations induced by their downstream task, objective function, and other parameters. The quality of the learned representations impacts the DNN's generalization ability and the coherence of the emerging latent space. The Information Bottleneck (IB) provides a hypothetically optimal framework for data modeling, yet it is often intractable. Recent efforts combined DNNs with the IB by applying VAE-inspired variational methods to approximate bounds on mutual information, resulting in improved robustness to adversarial attacks. This work introduces a new and tighter variational bound for the IB, improving performance of previous IB-inspired DNNs. These advancements strengthen the case for the IB and its variational approximations as a data modeling framework, and provide a simple method to significantly enhance the adversarial robustness of classifier DNNs.

Graph Neural Networks Use Graphs When They Shouldn't

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.

The Case Against Explainability

As artificial intelligence (AI) becomes more prevalent there is a growing demand from regulators to accompany decisions made by such systems with explanations. However, a persistent gap exists between the need to execute a meaningful right to explanation vs. the ability of Machine Learning systems to deliver on such a legal requirement. The regulatory appeal towards "a right to explanation" of AI systems can be attributed to the significant role of explanations, part of the notion called reason-giving, in law. Therefore, in this work we examine reason-giving's purposes in law to analyze whether reasons provided by end-user Explainability can adequately fulfill them. We find that reason-giving's legal purposes include: (a) making a better and more just decision, (b) facilitating due-process, (c) authenticating human agency, and (d) enhancing the decision makers' authority. Using this methodology, we demonstrate end-user Explainabilty's inadequacy to fulfil reason-giving's role in law, given reason-giving's functions rely on its impact over a human decision maker. Thus, end-user Explainability fails, or is unsuitable, to fulfil the first, second and third legal function. In contrast we find that end-user Explainability excels in the fourth function, a quality which raises serious risks considering recent end-user Explainability research trends, Large Language Models' capabilities, and the ability to manipulate end-users by both humans and machines. Hence, we suggest that in some cases the right to explanation of AI systems could bring more harm than good to end users. Accordingly, this study carries some important policy ramifications, as it calls upon regulators and Machine Learning practitioners to reconsider the widespread pursuit of end-user Explainability and a right to explanation of AI systems.

Graph Trees with Attention

When dealing with tabular data, models based on regression and decision trees are a popular choice due to the high accuracy they provide on such tasks and their ease of application as compared to other model classes. Yet, when it comes to graph-structure data, current tree learning algorithms do not provide tools to manage the structure of the data other than relying on feature engineering. In this work we address the above gap, and introduce Graph Trees with Attention (GTA), a new family of tree-based learning algorithms that are designed to operate on graphs. GTA leverages both the graph structure and the features at the vertices and employs an attention mechanism that allows decisions to concentrate on sub-structures of the graph. We analyze GTA models and show that they are strictly more expressive than plain decision trees. We also demonstrate the benefits of GTA empirically on multiple graph and node prediction benchmarks. In these experiments, GTA always outperformed other tree-based models and often outperformed other types of graph-learning algorithms such as Graph Neural Networks (GNNs) and Graph Kernels. Finally, we also provide an explainability mechanism for GTA, and demonstrate it can provide intuitive explanations.

Inherent Inconsistencies of Feature Importance

The black-box nature of modern machine learning techniques invokes a practical and ethical need for explainability. Feature importance aims to meet this need by assigning scores to features, so humans can understand their influence on predictions. Feature importance can be used to explain predictions under different settings: of the entire sample space or a specific instance; of model behavior, or the dependencies in the data themselves. However, in most cases thus far, each of these settings was studied in isolation. We attempt to develop a sound feature importance score framework by defining a small set of desired properties. Surprisingly, we prove an inconsistency theorem, showing that the expected properties cannot hold simultaneously. To overcome this difficulty, we propose the novel notion of re-partitioning the feature space into separable sets. Such sets are constructed to contain features that exhibit inter-set independence with respect to the target variable. We show that there exists a unique maximal partitioning into separable sets. Moreover, assigning scores to separable sets, instead of single features, unifies the results of commonly used feature importance scores and annihilates the inconsistencies we demonstrated.

Break your Bandit Routine with LSD Rewards: a Last Switch Dependent Analysis of Satiation and Seasonality

Motivated by the fact that humans like some level of unpredictability or novelty, and might therefore get quickly bored when interacting with a stationary policy, we introduce a novel non-stationary bandit problem, where the expected reward of an arm is fully determined by the time elapsed since the arm last took part in a switch of actions. Our model generalizes previous notions of delay-dependent rewards, and also relaxes most assumptions on the reward function. This enables the modeling of phenomena such as progressive satiation and periodic behaviours. Building upon the Combinatorial Semi-Bandits (CSB) framework, we design an algorithm and prove a bound on its regret with respect to the optimal non-stationary policy (which is NP-hard to compute). Similarly to previous works, our regret analysis is based on defining and solving an appropriate trade-off between approximation and estimation. Preliminary experiments confirm the superiority of our algorithm over both the oracle greedy approach and a vanilla CSB solver.

Robust Model Compression Using Deep Hypotheses

Machine Learning models should ideally be compact and robust. Compactness provides efficiency and comprehensibility whereas robustness provides resilience. Both topics have been studied in recent years but in isolation. Here we present a robust model compression scheme which is independent of model types: it can compress ensembles, neural networks and other types of models into diverse types of small models. The main building block is the notion of depth derived from robust statistics. Originally, depth was introduced as a measure of the centrality of a point in a sample such that the median is the deepest point. This concept was extended to classification functions which makes it possible to define the depth of a hypothesis and the median hypothesis. Algorithms have been suggested to approximate the median but they have been limited to binary classification. In this study, we present a new algorithm, the Multiclass Empirical Median Optimization (MEMO) algorithm that finds a deep hypothesis in multi-class tasks, and prove its correctness. This leads to our Compact Robust Estimated Median Belief Optimization (CREMBO) algorithm for robust model compression. We demonstrate the success of this algorithm empirically by compressing neural networks and random forests into small decision trees, which are interpretable models, and show that they are more accurate and robust than other comparable methods. In addition, our empirical study shows that our method outperforms Knowledge Distillation on DNN to DNN compression.

Marginal Contribution Feature Importance -- an Axiomatic Approach for The Natural Case

When training a predictive model over medical data, the goal is sometimes to gain insights about a certain disease. In such cases, it is common to use feature importance as a tool to highlight significant factors contributing to that disease. As there are many existing methods for computing feature importance scores, understanding their relative merits is not trivial. Further, the diversity of scenarios in which they are used lead to different expectations from the feature importance scores. While it is common to make the distinction between local scores that focus on individual predictions and global scores that look at the contribution of a feature to the model, another important division distinguishes model scenarios, in which the goal is to understand predictions of a given model from natural scenarios, in which the goal is to understand a phenomenon such as a disease. We develop a set of axioms that represent the properties expected from a feature importance function in the natural scenario and prove that there exists only one function that satisfies all of them, the Marginal Contribution Feature Importance (MCI). We analyze this function for its theoretical and empirical properties and compare it to other feature importance scores. While our focus is the natural scenario, we suggest that our axiomatic approach could be carried out in other scenarios too.

Algorithmic Rewriting of Health-Related Ads to Improve their Performance

Search advertising is one of the most commonly-used methods of advertising. Past work has shown that search advertising can be employed to improve health by eliciting positive behavioral change. However, writing effective advertisements requires expertise and (possible expensive) experimentation, both of which may not be available to public health authorities wishing to elicit such behavioral changes, especially when dealing with a public health crises such as epidemic outbreaks. Here we develop an algorithm which builds on past advertising data to train a sequence-to-sequence Deep Neural Network which "translates" advertisements into optimized ads that are more likely to be clicked. The network is trained using more than 114 thousands ads shown on Microsoft Advertising. We apply this translator to two health related domains: Medical Symptoms (MS) and Preventative Healthcare (PH) and measure the improvements in click-through rates (CTR). Our experiments show that the generated ads are predicted to have higher CTR in 81% of MS ads and 76% of PH ads. To understand the differences between the generated ads and the original ones we develop estimators for the affective attributes of the ads. We show that the generated ads contain more calls-to-action and that they reflect higher valence (36% increase) and higher arousal (87%) on a sample of 1000 ads. Finally, we run an advertising campaign where 10 random ads and their rephrased versions from each of the domains are run in parallel. We show an average improvement in CTR of 68% for the generated ads compared to the original ads. Our results demonstrate the ability to automatically optimize advertisement for the health domain. We believe that our work offers health authorities an improved ability to help nudge people towards healthier behaviors while saving the time and cost needed to optimize advertising campaigns.

The Automated Copywriter: Algorithmic Rephrasing of Health-Related Advertisements to Improve their Performance

Search advertising is one of the most commonly-used methods of advertising. Past work has shown that search advertising can be employed to improve health by eliciting positive behavioral change. However, writing effective advertisements requires expertise and (possible expensive) experimentation, both of which may not be available to public health authorities wishing to elicit such behavioral changes, especially when dealing with a public health crises such as epidemic outbreaks. Here we develop an algorithm which builds on past advertising data to train a sequence-to-sequence Deep Neural Network which "translates" advertisements into optimized ads that are more likely to be clicked. The network is trained using more than 114 thousands ads shown on Microsoft Advertising. We apply this translator to two health related domains: Medical Symptoms (MS) and Preventative Healthcare (PH) and measure the improvements in click-through rates (CTR). Our experiments show that the generated ads are predicted to have higher CTR in 81% of MS ads and 76% of PH ads. To understand the differences between the generated ads and the original ones we develop estimators for the affective attributes of the ads. We show that the generated ads contain more calls-to-action and that they reflect higher valence (36% increase) and higher arousal (87%) on a sample of 1000 ads. Finally, we run an advertising campaign where 10 random ads and their rephrased versions from each of the domains are run in parallel. We show an average improvement in CTR of 68% for the generated ads compared to the original ads. Our results demonstrate the ability to automatically optimize advertisement for the health domain. We believe that our work offers health authorities an improved ability to help nudge people towards healthier behaviors while saving the time and cost needed to optimize advertising campaigns.