We consider a reinforcement learning (RL) setting in which the agent interacts with a sequence of episodic MDPs. At the start of each episode the agent has access to some side-information or context that determines the dynamics of the MDP for that episode. Our setting is motivated by applications in healthcare where baseline measurements of a patient at the start of a treatment episode form the context that may provide information about how the patient might respond to treatment decisions. We propose algorithms for learning in such Contextual Markov Decision Processes (CMDPs) under an assumption that the unobserved MDP parameters vary smoothly with the observed context. We also give lower and upper PAC bounds under the smoothness assumption. Because our lower bound has an exponential dependence on the dimension, we consider a tractable linear setting where the context is used to create linear combinations of a finite set of MDPs. For the linear setting, we give a PAC learning algorithm based on KWIK learning techniques.
As a step towards developing zero-shot task generalization capabilities in reinforcement learning (RL), we introduce a new RL problem where the agent should learn to execute sequences of instructions after learning useful skills that solve subtasks. In this problem, we consider two types of generalizations: to previously unseen instructions and to longer sequences of instructions. For generalization over unseen instructions, we propose a new objective which encourages learning correspondences between similar subtasks by making analogies. For generalization over sequential instructions, we present a hierarchical architecture where a meta controller learns to use the acquired skills for executing the instructions. To deal with delayed reward, we propose a new neural architecture in the meta controller that learns when to update the subtask, which makes learning more efficient. Experimental results on a stochastic 3D domain show that the proposed ideas are crucial for generalization to longer instructions as well as unseen instructions.
This paper proposes a novel deep reinforcement learning (RL) architecture, called Value Prediction Network (VPN), which integrates model-free and model-based RL methods into a single neural network. In contrast to typical model-based RL methods, VPN learns a dynamics model whose abstract states are trained to make option-conditional predictions of future values (discounted sum of rewards) rather than of future observations. Our experimental results show that VPN has several advantages over both model-free and model-based baselines in a stochastic environment where careful planning is required but building an accurate observation-prediction model is difficult. Furthermore, VPN outperforms Deep Q-Network (DQN) on several Atari games even with short-lookahead planning, demonstrating its potential as a new way of learning a good state representation.
We introduce a novel repeated Inverse Reinforcement Learning problem: the agent has to act on behalf of a human in a sequence of tasks and wishes to minimize the number of tasks that it surprises the human by acting suboptimally with respect to how the human would have acted. Each time the human is surprised, the agent is provided a demonstration of the desired behavior by the human. We formalize this problem, including how the sequence of tasks is chosen, in a few different ways and provide some foundational results.
In cooperative multiagent planning, it can often be beneficial for an agent to make commitments about aspects of its behavior to others, allowing them in turn to plan their own behaviors without taking the agent's detailed behavior into account. Extending previous work in the Bayesian setting, we consider instead a worst-case setting in which the agent has a set of possible environments (MDPs) it could be in, and develop a commitment semantics that allows for probabilistic guarantees on the agent's behavior in any of the environments it could end up facing. Crucially, an agent receives observations (of reward and state transitions) that allow it to potentially eliminate possible environments and thus obtain higher utility by adapting its policy to the history of observations. We develop algorithms and provide theory and some preliminary empirical results showing that they ensure an agent meets its commitments with history-dependent policies while minimizing maximum regret over the possible environments.
In this paper, we introduce a new set of reinforcement learning (RL) tasks in Minecraft (a flexible 3D world). We then use these tasks to systematically compare and contrast existing deep reinforcement learning (DRL) architectures with our new memory-based DRL architectures. These tasks are designed to emphasize, in a controllable manner, issues that pose challenges for RL methods including partial observability (due to first-person visual observations), delayed rewards, high-dimensional visual observations, and the need to use active perception in a correct manner so as to perform well in the tasks. While these tasks are conceptually simple to describe, by virtue of having all of these challenges simultaneously they are difficult for current DRL architectures. Additionally, we evaluate the generalization performance of the architectures on environments not used during training. The experimental results show that our new architectures generalize to unseen environments better than existing DRL architectures.
Monte Carlo Tree Search (MCTS) methods have proven powerful in planning for sequential decision-making problems such as Go and video games, but their performance can be poor when the planning depth and sampling trajectories are limited or when the rewards are sparse. We present an adaptation of PGRD (policy-gradient for reward-design) for learning a reward-bonus function to improve UCT (a MCTS algorithm). Unlike previous applications of PGRD in which the space of reward-bonus functions was limited to linear functions of hand-coded state-action-features, we use PGRD with a multi-layer convolutional neural network to automatically learn features from raw perception as well as to adapt the non-linear reward-bonus function parameters. We also adopt a variance-reducing gradient method to improve PGRD's performance. The new method improves UCT's performance on multiple ATARI games compared to UCT without the reward bonus. Combining PGRD and Deep Learning in this way should make adapting rewards for MCTS algorithms far more widely and practically applicable than before.
We consider a setting for Inverse Reinforcement Learning (IRL) where the learner is extended with the ability to actively select multiple environments, observing an agent's behavior on each environment. We first demonstrate that if the learner can experiment with any transition dynamics on some fixed set of states and actions, then there exists an algorithm that reconstructs the agent's reward function to the fullest extent theoretically possible, and that requires only a small (logarithmic) number of experiments. We contrast this result to what is known about IRL in single fixed environments, namely that the true reward function is fundamentally unidentifiable. We then extend this setting to the more realistic case where the learner may not select any transition dynamic, but rather is restricted to some fixed set of environments that it may try. We connect the problem of maximizing the information derived from experiments to submodular function maximization and demonstrate that a greedy algorithm is near optimal (up to logarithmic factors). Finally, we empirically validate our algorithm on an environment inspired by behavioral psychology.
Motivated by vision-based reinforcement learning (RL) problems, in particular Atari games from the recent benchmark Aracade Learning Environment (ALE), we consider spatio-temporal prediction problems where future (image-)frames are dependent on control variables or actions as well as previous frames. While not composed of natural scenes, frames in Atari games are high-dimensional in size, can involve tens of objects with one or more objects being controlled by the actions directly and many other objects being influenced indirectly, can involve entry and departure of objects, and can involve deep partial observability. We propose and evaluate two deep neural network architectures that consist of encoding, action-conditional transformation, and decoding layers based on convolutional neural networks and recurrent neural networks. Experimental results show that the proposed architectures are able to generate visually-realistic frames that are also useful for control over approximately 100-step action-conditional futures in some games. To the best of our knowledge, this paper is the first to make and evaluate long-term predictions on high-dimensional video conditioned by control inputs.
In this work, we introduce graphical modelsfor multi-player game theory, and give powerful algorithms for computing their Nash equilibria in certain cases. An n-player game is given by an undirected graph on n nodes and a set of n local matrices. The interpretation is that the payoff to player i is determined entirely by the actions of player i and his neighbors in the graph, and thus the payoff matrix to player i is indexed only by these players. We thus view the global n-player game as being composed of interacting local games, each involving many fewer players. Each player's action may have global impact, but it occurs through the propagation of local influences.Our main technical result is an efficient algorithm for computing Nash equilibria when the underlying graph is a tree (or can be turned into a tree with few node mergings). The algorithm runs in time polynomial in the size of the representation (the graph and theassociated local game matrices), and comes in two related but distinct flavors. The first version involves an approximation step, and computes a representation of all approximate Nash equilibria (of which there may be an exponential number in general). The second version allows the exact computation of Nash equilibria at the expense of weakened complexity bounds. The algorithm requires only local message-passing between nodes (and thus can be implemented by the players themselves in a distributed manner). Despite an analogy to inference in Bayes nets that we develop, the analysis of our algorithm is more involved than that for the polytree algorithm in, owing partially to the fact that we must either compute, or select from, an exponential number of potential solutions. We discuss a number of extensions, such as the computation of equilibria with desirable global properties (e.g. maximizing global return), and directions for further research.