Reinforcement Learning aims at identifying and evaluating efficient control policies from data. In many real-world applications, the learner is not allowed to experiment and cannot gather data in an online manner (this is the case when experimenting is expensive, risky or unethical). For such applications, the reward of a given policy (the target policy) must be estimated using historical data gathered under a different policy (the behavior policy). Most methods for this learning task, referred to as Off-Policy Evaluation (OPE), do not come with accuracy and certainty guarantees. We present a novel OPE method based on Conformal Prediction that outputs an interval containing the true reward of the target policy with a prescribed level of certainty. The main challenge in OPE stems from the distribution shift due to the discrepancies between the target and the behavior policies. We propose and empirically evaluate different ways to deal with this shift. Some of these methods yield conformalized intervals with reduced length compared to existing approaches, while maintaining the same certainty level.
We investigate the sample complexity of learning the optimal arm for multi-task bandit problems. Arms consist of two components: one that is shared across tasks (that we call representation) and one that is task-specific (that we call predictor). The objective is to learn the optimal (representation, predictor)-pair for each task, under the assumption that the optimal representation is common to all tasks. Within this framework, efficient learning algorithms should transfer knowledge across tasks. We consider the best-arm identification problem for a fixed confidence, where, in each round, the learner actively selects both a task, and an arm, and observes the corresponding reward. We derive instance-specific sample complexity lower bounds satisfied by any $(\delta_G,\delta_H)$-PAC algorithm (such an algorithm identifies the best representation with probability at least $1-\delta_G$, and the best predictor for a task with probability at least $1-\delta_H$). We devise an algorithm OSRL-SC whose sample complexity approaches the lower bound, and scales at most as $H(G\log(1/\delta_G)+ X\log(1/\delta_H))$, with $X,G,H$ being, respectively, the number of tasks, representations and predictors. By comparison, this scaling is significantly better than the classical best-arm identification algorithm that scales as $HGX\log(1/\delta)$.
* Accepted at the Thirty-Seventh AAAI Conference on Artificial
Non-differentiable controllers and rule-based policies are widely used for controlling real systems such as robots and telecommunication networks. In this paper, we present a practical reinforcement learning method which improves upon such existing policies with a model-based approach for better sample efficiency. Our method significantly outperforms state-of-the-art model-based methods, in terms of sample efficiency, on several widely used robotic benchmark tasks. We also demonstrate the effectiveness of our approach on a control problem in the telecommunications domain, where model-based methods have not previously been explored. Experimental results indicate that a strong initial performance can be achieved and combined with improved sample efficiency. We further motivate the design of our algorithm with a theoretical lower bound on the performance.
A recent body of literature has investigated the effect of data poisoning attacks on data-driven control methods. Data poisoning attacks are well-known to the Machine Learning community, which, however, make use of assumptions, such as cross-sample independence, that in general do not hold for dynamical systems. As a consequence, attacks, and detection methods, operate differently from the i.i.d. setting studied in classical supervised problems. In particular, data poisoning attacks against data-driven control methods can be fundamentally seen as changing the behavior of the dynamical system described by the data. In this work, we study this phenomenon through the lens of statistical testing, and verify the detectability of different attacks for a linear dynamical system. On the basis of the arguments hereby presented, we propose a stealthy data poisoning attack that can escape classical detection tests, and conclude by showing the efficiency of the proposed attack.
We investigate the problem of designing optimal stealthy poisoning attacks on the control channel of Markov decision processes (MDPs). This research is motivated by the recent interest of the research community for adversarial and poisoning attacks applied to MDPs, and reinforcement learning (RL) methods. The policies resulting from these methods have been shown to be vulnerable to attacks perturbing the observations of the decision-maker. In such an attack, drawing inspiration from adversarial examples used in supervised learning, the amplitude of the adversarial perturbation is limited according to some norm, with the hope that this constraint will make the attack imperceptible. However, such constraints do not grant any level of undetectability and do not take into account the dynamic nature of the underlying Markov process. In this paper, we propose a new attack formulation, based on information-theoretical quantities, that considers the objective of minimizing the detectability of the attack as well as the performance of the controlled process. We analyze the trade-off between the efficiency of the attack and its detectability. We conclude with examples and numerical simulations illustrating this trade-off.
Recent successes in the Machine Learning community have led to a steep increase in the number of papers submitted to conferences. This increase made more prominent some of the issues that affect the current review process used by these conferences. The review process has several issues that may undermine the nature of scientific research, which is of being fully objective, apolitical, unbiased and free of misconduct (such as plagiarism, cheating, improper influence, and other improprieties). In this work, we study the problem of reviewers' recruitment, infringements of the double-blind process, fraudulent behaviors, biases in numerical ratings, and the appendix phenomenon (i.e., the fact that it is becoming more common to publish results in the appendix section of a paper). For each of these problems, we provide a short description and possible solutions. The goal of this work is to raise awareness in the Machine Learning community regarding these issues.
Control policies, trained using the Deep Reinforcement Learning, have been recently shown to be vulnerable to adversarial attacks introducing even very small perturbations to the policy input. The attacks proposed so far have been designed using heuristics, and build on existing adversarial example crafting techniques used to dupe classifiers in supervised learning. In contrast, this paper investigates the problem of devising optimal attacks, depending on a well-defined attacker's objective, e.g., to minimize the main agent average reward. When the policy and the system dynamics, as well as rewards, are known to the attacker, a scenario referred to as a white-box attack, designing optimal attacks amounts to solving a Markov Decision Process. For what we call black-box attacks, where neither the policy nor the system is known, optimal attacks can be trained using Reinforcement Learning techniques. Through numerical experiments, we demonstrate the efficiency of our attacks compared to existing attacks (usually based on Gradient methods). We further quantify the potential impact of attacks and establish its connection to the smoothness of the policy under attack. Smooth policies are naturally less prone to attacks (this explains why Lipschitz policies, with respect to the state, are more resilient). Finally, we show that from the main agent perspective, the system uncertainties and the attacker can be modeled as a Partially Observable Markov Decision Process. We actually demonstrate that using Reinforcement Learning techniques tailored to POMDP (e.g. using Recurrent Neural Networks) leads to more resilient policies.