Learning how to act when there are many available actions in each state is a challenging task for Reinforcement Learning (RL) agents, especially when many of the actions are redundant or irrelevant. In such cases, it is sometimes easier to learn which actions not to take. In this work, we propose the Action-Elimination Deep Q-Network (AE-DQN) architecture that combines a Deep RL algorithm with an Action Elimination Network (AEN) that eliminates sub-optimal actions. The AEN is trained to predict invalid actions, supervised by an external elimination signal provided by the environment. Simulations demonstrate a considerable speedup and added robustness over vanilla DQN in text-based games with over a thousand discrete actions.

Recent work in imitation learning articulate their formulation around the GAIL architecture, relying on the adversarial training procedure introduced in GANs. Albeit successful at generating behaviours similar to those demonstrated to the agent, GAIL suffers from a high sample complexity in the number of interactions it has to carry out in the environment in order to achieve satisfactory performance. In this work, we dramatically shrink the amount of interactions with the environment by leveraging an off-policy actor-critic architecture. Additionally, employing deterministic policy gradients allows us to treat the learned reward as a differentiable node in the computational graph, while preserving the model-free nature of our approach. Our experiments span a variety of continuous control tasks.

Though successful in high-dimensional domains, deep reinforcement learning exhibits high sample complexity and suffers from stability issues as reported by researchers and practitioners in the field. These problems hinder the application of such algorithms in real-world and safety-critical scenarios. In this paper, we take steps towards stable and efficient reinforcement learning by following a model-based approach that is known to reduce agent-environment interactions. Namely, our method augments deep Q-networks (DQNs) with model predictions for transitions, rewards, and termination flags. Having the model at hand, we then conduct a rigorous theoretical study of our algorithm and show, for the first time, convergence to a stationary point. En route, we provide a counter-example showing that 'vanilla' DQNs can diverge confirming practitioners' and researchers' experiences. Our proof is novel in its own right and can be extended to other forms of deep reinforcement learning. In particular, we believe exploiting the relation between reinforcement (with deep function approximators) and online learning can serve as a recipe for future proofs in the domain. Finally, we validate our theoretical results in 20 games from the Atari benchmark. Our results show that following the proposed model-based learning approach not only ensures convergence but leads to a reduction in sample complexity and superior performance.

We consider the problem of learning optimal policies from observational data in a way that satisfies certain fairness criteria. The issue of fairness arises where some covariates used in decision making are sensitive features, or are correlated with sensitive features. (Nabi and Shpitser 2018) formalized fairness in the context of regression problems as constraining the causal effects of sensitive features along certain disallowed causal pathways. The existence of these causal effects may be called retrospective unfairness in the sense of already being present in the data before analysis begins, and may be due to discriminatory practices or the biased way in which variables are defined or recorded. In the context of learning policies, what we call prospective bias, i.e., the inappropriate dependence of learned policies on sensitive features, is also possible. In this paper, we use methods from causal and semiparametric inference to learn optimal policies in a way that addresses both retrospective bias in the data, and prospective bias due to the policy. In addition, our methods appropriately address statistical bias due to model misspecification and confounding bias, which are important in the estimation of path-specific causal effects from observational data. We apply our methods to both synthetic data and real criminal justice data.

Inverse reinforcement learning (IRL) attempts to infer human rewards or preferences from observed behavior. Since human planning systematically deviates from rationality, several approaches have been tried to account for specific human shortcomings. However, there has been little analysis of the general problem of inferring the reward of a human of unknown rationality. The observed behavior can, in principle, be decomposed into two components: a reward function and a planning algorithm, both of which have to be inferred from behavior. This paper presents a No Free Lunch theorem, showing that, without making `normative' assumptions beyond the data, nothing about the human reward function can be deduced from human behavior. Unlike most No Free Lunch theorems, this cannot be alleviated by regularising with simplicity assumptions. We show that the simplest hypotheses which explain the data are generally degenerate.

Research in deep reinforcement learning (RL) has coalesced around improving performance on benchmarks like the Arcade Learning Environment. However, these benchmarks conspicuously miss important characteristics like abrupt context-dependent shifts in strategy and temporal sensitivity that are often present in real-world domains. As a result, RL research has not focused on these challenges, resulting in algorithms which do not understand critical changes in context, and have little notion of real world time. To tackle this issue, this paper introduces the game of Space Fortress as a RL benchmark which incorporates these characteristics. We show that existing state-of-the-art RL algorithms are unable to learn to play the Space Fortress game. We then confirm that this poor performance is due to the RL algorithms' context insensitivity and reward sparsity. We also identify independent axes along which to vary context and temporal sensitivity, allowing Space Fortress to be used as a testbed for understanding both characteristics in combination and also in isolation. We release Space Fortress as an open-source Gym environment.

Finite-horizon lookahead policies are abundantly used in Reinforcement Learning and demonstrate impressive empirical success. Usually, the lookahead policies are implemented with specific planning methods such as Monte Carlo Tree Search (e.g. in AlphaZero). Referring to the planning problem as tree search, a reasonable practice in these implementations is to back up the value only at the leaves while the information obtained at the root is not leveraged other than for updating the policy. Here, we question the potency of this approach.Namely, the latter procedure is non-contractive in general, and its convergence is not guaranteed. Our proposed enhancement is straightforward and simple: use the return from the optimal tree path to back up the values at the descendants of the root. This leads to a \gamma^h-contracting procedure, where \gamma is the discount factor and $h$ is the tree depth. To establish our results, we first introduce a notion called multiple-step greedy consistency. We then provide convergence rates for two algorithmic instantiations of the above enhancement in the presence of noise injected to both the tree search stage and value estimation stage.

Experience replay is an important technique for addressing sample-inefficiency in deep reinforcement learning (RL), but faces difficulty in learning from binary and sparse rewards due to disproportionately few successful experiences in the replay buffer. Hindsight experience replay (HER) (Andrychowicz et al. 2017) was recently proposed to tackle this difficulty by manipulating unsuccessful transitions, but in doing so, HER introduces a significant bias in the replay buffer experiences and therefore achieves a suboptimal improvement in sample-efficiency. In this paper, we present an analysis on the source of bias in HER, and propose a simple and effective method to counter the bias, to most effectively harness the sample-efficiency provided by HER. Our method, motivated by counterfactual reasoning and called ARCHER, extends HER with a tradeoff to make rewards calculated for hindsight experiences numerically greater than real rewards. We validate our algorithm on two continuous control environments from DeepMind Control Suite (Tassa et al. 2018) - Reacher and Finger, which simulate manipulation tasks with a robotic arm - in combination with various reward functions, task complexities and goal sampling strategies. Our experiments consistently demonstrate that countering bias using more aggressive hindsight rewards increases sample efficiency, thus establishing the greater benefit of ARCHER in RL applications with limited computing budget.

This work provides a thorough study on how reward scaling can affect performance of deep reinforcement learning agents. In particular, we would like to answer the question that How does reward scaling affect non-saturating ReLU networks in RL? This question matters because ReLU is one of the most effective activation functions for deep learning models. We also propose an Adaptive Network Scaling framework to find a suitable scale of the rewards during learning for better performance. We conducted empirical studies to justify the solution.

Real-time bidding (RTB) is an important mechanism in online display advertising, where a proper bid for each page view plays an essential role for good marketing results. Budget constrained bidding is a typical scenario in RTB where the advertisers hope to maximize the total value of the winning impressions under a pre-set budget constraint. However, the optimal bidding strategy is hard to be derived due to the complexity and volatility of the auction environment. To address these challenges, in this paper, we formulate budget constrained bidding as a Markov Decision Process and propose a model-free reinforcement learning framework to resolve the optimization problem. Our analysis shows that the immediate reward from environment is misleading under a critical resource constraint. Therefore, we innovate a reward function design methodology for the reinforcement learning problems with constraints. Based on the new reward design, we employ a deep neural network to learn the appropriate reward so that the optimal policy can be learned effectively. Different from the prior model-based work, which suffers from the scalability problem, our framework is easy to be deployed in large-scale industrial applications. The experimental evaluations demonstrate the effectiveness of our framework on large-scale real datasets.