Normally when comparing different models/algorithms, effort should be taken to find the best performing parameters (or more generally most suitable formalisations) for each model.

In the first-order query model for zero-sum K\times K matrix games, players observe the expected pay-offs for all their possible actions under the randomized action played by their opponent. Surprisingly, the optimal number of such queries, as a function of both \epsilon and K, is not known. We make progress on this question on two fronts. First, we fully characterise the query complexity of learning exact equilibria ( \epsilon 0), by showing that they require a number of queries that is linear in K, which means that it is essentially as hard as querying the whole matrix, which can also be done with K queries. We argue that, unfortunately, obtaining a matching lower bound is not possible with existing techniques: we prove that no lower bound can be derived by constructing hard matrices whose entries take values in a known countable set, because such matrices can be fully identified by a single query.

Research in machine learning is making progress in fixing its own reproducibility crisis. Reinforcement learning (RL), in particular, faces its own set of unique challenges. Comparison of point estimates, and plots that show successful convergence to the optimal policy during training, may obfuscate overfitting or dependence on the experimental setup. Although researchers in RL have proposed reliability metrics that account for uncertainty to better understand each algorithm's strengths and weaknesses, the recommendations of past work do not assume the presence of out-of-distribution observations. We propose a set of evaluation methods that measure the robustness of RL algorithms under distribution shifts. The tools presented here argue for the need to account for performance over time while the agent is acting in its environment. In particular, we recommend time series analysis as a method of observational RL evaluation. We also show that the unique properties of RL and simulated dynamic environments allow us to make stronger assumptions to justify the measurement of causal impact in our evaluations. We then apply these tools to single-agent and multi-agent environments to show the impact of introducing distribution shifts during test time. We present this methodology as a first step toward rigorous RL evaluation in the presence of distribution shifts.

Devising procedures for downstream task-oriented generative model selections is an unresolved problem of practical importance. Existing studies focused on the utility of a single family of generative models. They provided limited insights on how synthetic data practitioners select the best family generative models for synthetic training tasks given a specific combination of machine learning model class and performance metric. In this paper, we approach the downstream task-oriented generative model selections problem in the case of training fraud detection models and investigate the best practice given different combinations of model interpretability and model performance constraints. Our investigation supports that, while both Neural Network(NN)-based and Bayesian Network(BN)-based generative models are both good to complete synthetic training task under loose model interpretability constrain, the BN-based generative models is better than NN-based when synthetic training fraud detection model under strict model interpretability constrain. Our results provides practical guidance for machine learning practitioner who is interested in replacing their training dataset from real to synthetic, and shed lights on more general downstream task-oriented generative model selection problems.

Generative adversarial neural networks (GANs) [1] [2] [3] [4] are deep learning (DL) models whereby a dataset is used by an agent, called the generator, to sample white noise from a latent space and simulate a data distribution to create new (fake) data that resemble the original data it has been trained on. Another agent, called the discriminator, has to correctly discern between the original data (provided by the external environment for training) and the fake data (produced by the generator). The generator prevails over the discriminator if the latter does not succeed in distinguishing anymore the original from the fake. The discriminator prevails over the generator if the fake data created by the generator is categorized as fake, and the original data is still categorized as original. An illustration that describes a GANs model is shown in Figure 1.

Adversarial robustness has proven to be a required property of machine learning algorithms. A key and often overlooked aspect of this problem is to try to make the adversarial noise magnitude as large as possible to enhance the benefits of the model robustness. We show that the well-established algorithm called "adversarial training" fails to train a deep neural network given a large, but reasonable, perturbation magnitude. In this paper, we propose a simple yet effective initialization of the network weights that makes learning on higher levels of noise possible. We next evaluate this idea rigorously on MNIST ($\epsilon$ up to $\approx 0.40$) and CIFAR10 ($\epsilon$ up to $\approx 32/255$) datasets assuming the $\ell_{\infty}$ attack model. Additionally, in order to establish the limits of $\epsilon$ in which the learning is feasible, we study the optimal robust classifier assuming full access to the joint data and label distribution. Then, we provide some theoretical results on the adversarial accuracy for a simple multi-dimensional Bernoulli distribution, which yields some insights on the range of feasible perturbations for the MNIST dataset.

In this work we formally define the notions of adversarial perturbations, adversarial risk and adversarial training and analyze their properties. Our analysis provides several interesting insights into adversarial risk, adversarial training, and their relation to the classification risk, "traditional" training. We also show that adversarial training can result in models with better classification accuracy and can result in better explainable models than traditional training. Although adversarial training is computationally expensive, our results and insights suggest that one should prefer adversarial training over traditional risk minimization for learning complex models from data.

Thompson sampling has emerged as an effective heuristic for a broad range of online decision problems. In its basic form, the algorithm requires computing and sampling from a posterior distribution over models, which is tractable only for simple special cases. This paper develops ensemble sampling, which aims to approximate Thompson sampling while maintaining tractability even in the face of complex models such as neural networks. Ensemble sampling dramatically expands on the range of applications for which Thompson sampling is viable. We establish a theoretical basis that supports the approach and present computational results that offer further insight.