We study online agnostic learning, a problem that arises in episodic multi-agent reinforcement learning where the actions of the opponents are unobservable. We show that in this challenging setting, achieving sublinear regret against the best response in hindsight is statistically hard. We then consider a weaker notion of regret, and present an algorithm that achieves after $K$ episodes a sublinear $\tilde{\mathcal{O}}(K^{3/4})$ regret. This is the first sublinear regret bound (to our knowledge) in the online agnostic setting. Importantly, our regret bound is independent of the size of the opponents' action spaces. As a result, even when the opponents' actions are fully observable, our regret bound improves upon existing analysis (e.g., (Xie et al., 2020)) by an exponential factor in the number of opponents.

Despite empirical success, the theory of reinforcement learning (RL) with value function approximation remains fundamentally incomplete. Prior work has identified a variety of pathological behaviours that arise in RL algorithms that combine approximate on-policy evaluation and greedification. One prominent example is policy oscillation, wherein an algorithm may cycle indefinitely between policies, rather than converging to a fixed point. What is not well understood however is the quality of the policies in the region of oscillation. In this paper we present simple examples illustrating that in addition to policy oscillation and multiple fixed points -- the same basic issue can lead to convergence to the worst possible policy for a given approximation. Such behaviours can arise when algorithms optimize evaluation accuracy weighted by the distribution of states that occur under the current policy, but greedify based on the value of states which are rare or nonexistent under this distribution. This means the values used for greedification are unreliable and can steer the policy in undesirable directions. Our observation that this can lead to the worst possible policy shows that in a general sense such algorithms are unreliable. The existence of such examples helps to narrow the kind of theoretical guarantees that are possible and the kind of algorithmic ideas that are likely to be helpful. We demonstrate analytically and experimentally that such pathological behaviours can impact a wide range of RL and dynamic programming algorithms; such behaviours can arise both with and without bootstrapping, and with linear function approximation as well as with more complex parameterized functions like neural networks.

First-person object-interaction tasks in high-fidelity, 3D, simulated environments such as the AI2Thor virtual home-environment pose significant sample-efficiency challenges for reinforcement learning (RL) agents learning from sparse task rewards. To alleviate these challenges, prior work has provided extensive supervision via a combination of reward-shaping, ground-truth object-information, and expert demonstrations. In this work, we show that one can learn object-interaction tasks from scratch without supervision by learning an attentive object-model as an auxiliary task during task learning with an object-centric relational RL agent. Our key insight is that learning an object-model that incorporates object-attention into forward prediction provides a dense learning signal for unsupervised representation learning of both objects and their relationships. This, in turn, enables faster policy learning for an object-centric relational RL agent. We demonstrate our agent by introducing a set of challenging object-interaction tasks in the AI2Thor environment where learning with our attentive object-model is key to strong performance. Specifically, we compare our agent and relational RL agents with alternative auxiliary tasks to a relational RL agent equipped with ground-truth object-information, and show that learning with our object-model best closes the performance gap in terms of both learning speed and maximum success rate. Additionally, we find that incorporating object-attention into an object-model's forward predictions is key to learning representations which capture object-category and object-state.

In an ever expanding set of research and application areas, deep neural networks (DNNs) set the bar for algorithm performance. However, depending upon additional constraints such as processing power and execution time limits, or requirements such as verifiable safety guarantees, it may not be feasible to actually use such high-performing DNNs in practice. Many techniques have been developed in recent years to compress or distill complex DNNs into smaller, faster or more understandable models and controllers. This work seeks to provide a quantitative framework with metrics to systematically evaluate the outcome of such conversion processes, and identify reduced models that not only preserve a desired performance level, but also, for example, succinctly explain the latent knowledge represented by a DNN. We illustrate the effectiveness of the proposed approach on the evaluation of decision tree variants in the context of benchmark reinforcement learning tasks.

Model-based reinforcement learning methods learn a dynamics model with real data sampled from the environment and leverage it to generate simulated data to derive an agent. However, due to the potential distribution mismatch between simulated data and real data, this could lead to degraded performance. Despite much effort being devoted to reducing this distribution mismatch, existing methods fail to solve it explicitly. In this paper, we investigate how to bridge the gap between real and simulated data due to inaccurate model estimation for better policy optimization. To begin with, we first derive a lower bound of the expected return, which naturally inspires a bound maximization algorithm by aligning the simulated and real data distributions. To this end, we propose a novel model-based reinforcement learning framework AMPO, which introduces unsupervised model adaptation to minimize the integral probability metric (IPM) between feature distributions from real and simulated data. Instantiating our framework with Wasserstein-1 distance gives a practical model-based approach. Empirically, our approach achieves state-of-the-art performance in terms of sample efficiency on a range of continuous control benchmark tasks.

To operate successfully in unstructured open-world environments, autonomous intelligent agents need to solve many different tasks and learn new tasks quickly. Reinforcement learning has enabled artificial agents to solve complex tasks both in simulation and real world. However, it requires collecting large amounts of experience in the environment, and the agent learns only that particular task, much like a student memorizing a lecture without understanding. Self-supervised reinforcement learning has emerged as an alternative, where the agent only follows an intrinsic objective that is independent of any individual task, analogously to unsupervised representation learning. After experimenting with the environment without supervision, the agent builds an understanding of the environment, which enables it to adapt to specific downstream tasks more efficiently.

We evaluate the ability of temporal difference learning to track the reward function of a policy as it changes over time. Our results apply a new adiabatic theorem that bounds the mixing time of time-inhomogeneous Markov chains. We derive finite-time bounds for tabular temporal difference learning and $Q$-learning when the policy used for training changes in time. To achieve this, we develop bounds for stochastic approximation under asynchronous adiabatic updates.

Generating asset-specific trading signals based on the financial conditions of the assets is one of the challenging problems in automated trading. Various asset trading rules are proposed experimentally based on different technical analysis techniques. However, these kind of trading strategies are profitable, extracting new asset-specific trading rules from vast historical data to increase total return and decrease the risk of portfolios is difficult for human experts. Recently, various deep reinforcement learning (DRL) methods are employed to learn the new trading rules for each asset. In this paper, a novel DRL model with various feature extraction modules is proposed. The effect of different input representations on the performance of the models is investigated and the performance of DRL-based models in different markets and asset situations is studied. The proposed model in this work outperformed the other state-of-the-art models in learning single asset-specific trading rules and obtained a total return of almost 262% in two years on a specific asset while the best state-of-the-art model get 78% on the same asset in the same time period.

Deep reinforcement learning has recently shown many impressive successes. However, one major obstacle towards applying such methods to real-world problems is their lack of data-efficiency. To this end, we propose the Bottleneck Simulator: a model-based reinforcement learning method which combines a learned, factorized transition model of the environment with rollout simulations to learn an effective policy from few examples. The learned transition model employs an abstract, discrete (bottleneck) state, which increases sample efficiency by reducing the number of model parameters and by exploiting structural properties of the environment. We provide a mathematical analysis of the Bottleneck Simulator in terms of fixed points of the learned policy, which reveals how performance is affected by four distinct sources of error: an error related to the abstract space structure, an error related to the transition model estimation variance, an error related to the transition model estimation bias, and an error related to the transition model class bias. Finally, we evaluate the Bottleneck Simulator on two natural language processing tasks: a text adventure game and a real-world, complex dialogue response selection task.

Action-value estimation is a critical component of many reinforcement learning (RL) methods whereby sample complexity relies heavily on how fast a good estimator for action value can be learned. By viewing this problem through the lens of representation learning, good representations of both state and action can facilitate action-value estimation. While advances in deep learning have seamlessly driven progress in learning state representations, given the specificity of the notion of agency to RL, little attention has been paid to learning action representations. We conjecture that leveraging the combinatorial structure of multi-dimensional action spaces is a key ingredient for learning good representations of action. To test this, we set forth the action hypergraph networks framework---a class of functions for learning action representations with a relational inductive bias. Using this framework we realise an agent class based on a combination with deep Q-networks, which we dub hypergraph Q-networks. We show the effectiveness of our approach on a myriad of domains: illustrative prediction problems under minimal confounding effects, Atari 2600 games, and physical control benchmarks.