In recent years, research has shown that transfer learning methods can be leveraged to construct curricula that sequence a series of simpler tasks such that performance on a final target task is improved. A major limitation of existing approaches is that such curricula are handcrafted by humans that are typically domain experts. To address this limitation, we introduce a method to generate a curriculum based on task descriptors and a novel metric of transfer potential. Our method automatically generates a curriculum as a directed acyclic graph (as opposed to a linear sequence as done in existing work). Experiments in both discrete and continuous domains show that our method produces curricula that improve the agent's learning performance when compared to the baseline condition of learning on the target task from scratch.

In order for reinforcement learning systems to learn quickly in vast action spaces such as the space of all possible pieces of text or the space of all images, leveraging human intuition and creativity is key. However, a human-designed action space is likely to be initially imperfect and limited; furthermore, humans may improve at creating useful actions with practice or new information. Therefore, we propose a framework in which a human adds actions to a reinforcement learning system over time to boost performance. In this setting, however, it is key that we use human effort as efficiently as possible, and one significant danger is that humans waste effort adding actions at places (states) that aren't very important. Therefore, we propose Expected Local Improvement (ELI), an automated method which selects states at which to query humans for a new action. We evaluate ELI on a variety of simulated domains adapted from the literature, including domains with over a million actions and domains where the simulated experts change over time. We find ELI demonstrates excellent empirical performance, even in settings where the synthetic "experts" are quite poor.

This article addresses a particular Transfer Reinforcement Learning (RL) problem: when dynamics do not change from one task to another, and only the reward function does. Our method relies on two ideas, the first one is that transition samples obtained from a task can be reused to learn on any other task: an immediate reward estimator is learnt in a supervised fashion and for each sample, the reward entry is changed by its reward estimate. The second idea consists in adopting the optimism in the face of uncertainty principle and to use upper bound reward estimates. Our method is tested on a navigation task, under four Transfer RL experimental settings: with a known reward function, with strong and weak expert knowledge on the reward function, and with a completely unknown reward function. It is also evaluated in a Multi-Task RL experiment and compared with the state-of-the-art algorithms. Results reveal that this method constitutes a major improvement for transfer/multi-task problems that share dynamics.

Advances in deep reinforcement learning have allowed autonomous agents to perform well on Atari games, often outperforming humans, using only raw pixels to make their decisions. However, most of these games take place in 2D environments that are fully observable to the agent. In this paper, we present the first architecture to tackle 3D environments in first-person shooter games, that involve partially observable states. Typically, deep reinforcement learning methods only utilize visual input for training. We present a method to augment these models to exploit game feature information such as the presence of enemies or items, during the training phase. Our model is trained to simultaneously learn these features along with minimizing a Q-learning objective, which is shown to dramatically improve the training speed and performance of our agent. Our architecture is also modularized to allow different models to be independently trained for different phases of the game. We show that the proposed architecture substantially outperforms built-in AI agents of the game as well as average humans in deathmatch scenarios.

One of the long standing goals of Artificial Intelligence (AI) is to build cognitive agents which can perform complex tasks from raw sensory inputs without explicit supervision. Recent progress in combining Reinforcement Learning objective functions and Deep Learning architectures has achieved promising results for such tasks. An important aspect of such sequential decision making problems, which has largely been neglected, is for the agent to decide on the duration of time for which to commit to actions. Such action repetition is important for computational efficiency, which is necessary for the agent to respond in real-time to events (in applications such as self-driving cars). Action Repetition arises naturally in real life as well as simulated environments. The time scale of executing an action enables an agent (both humans and AI) to decide the granularity of control during task execution. Current state of the art Deep Reinforcement Learning models, whether they are off-policy or on-policy, consist of a framework with a static action repetition paradigm, wherein the action decided by the agent is repeated for a fixed number of time steps regardless of the contextual state while executing the task. In this paper, we propose a new framework - Dynamic Action Repetition which changes Action Repetition Rate (the time scale of repeating an action) from a hyper-parameter of an algorithm to a dynamically learnable quantity. At every decision-making step, our models allow the agent to commit to an action and the time scale of executing the action. We show empirically that such a dynamic time scale mechanism improves the performance on relatively harder games in the Atari 2600 domain, independent of the underlying Deep Reinforcement Learning algorithm used.

Policy search reinforcement learning (RL) allows agents to learn autonomously with limited feedback. However, such methods typically require extensive experience for successful behavior due to their tabula rasa nature. Multitask RL is an approach, which aims to reduce data requirements by allowing knowledge transfer between tasks. Although successful, current multitask learning methods suffer from scalability issues when considering large number of tasks. The main reasons behind this limitation is the reliance on centralized solutions. This paper proposes to a novel distributed multitask RL framework, improving the scalability across many different types of tasks. Our framework maps multitask RL to an instance of general consensus and develops an efficient decentralized solver. We justify the correctness of the algorithm both theoretically and empirically: we first proof an improvement of convergence speed to an order of O(1/k) with k being the number of iterations, and then show our algorithm surpassing others on multiple dynamical system benchmarks.

This paper is about the estimation of the maximum expected value of an infinite set of random variables.This estimation problem is relevant in many fields, like the Reinforcement Learning (RL) one.In RL it is well known that, in some stochastic environments, a bias in the estimation error can increase step-by-step the approximation error leading to large overestimates of the true action values. Recently, some approaches have been proposed to reduce such bias in order to get better action-value estimates, but are limited to finite problems.In this paper, we leverage on the recently proposed weighted estimator and on Gaussian process regression to derive a new method that is able to natively handle infinitely many random variables.We show how these techniques can be used to face both continuous state and continuous actions RL problems.To evaluate the effectiveness of the proposed approach we perform empirical comparisons with related approaches.

Policy gradient methods have been widely applied in reinforcement learning. For reasons of safety and cost, learning is often conducted using a simulator. However, learning in simulation does not traditionally utilise the opportunity to improve learning by adjusting certain environment variables - state features that are randomly determined by the environment in a physical setting but controllable in a simulator. Exploiting environment variables is crucial in domains containing significant rare events (SREs), e.g., unusual wind conditions that can crash a helicopter, which are rarely observed under random sampling but have a considerable impact on expected return. We propose off environment reinforcement learning (OFFER), which addresses such cases by simultaneously optimising the policy and a proposal distribution over environment variables. We prove that OFFER converges to a locally optimal policy and show experimentally that it learns better and faster than a policy gradient baseline.

Temporal abstraction is key to scaling up learning and planning in reinforcement learning. While planning with temporally extended actions is well understood, creating such abstractions autonomously from data has remained challenging.We tackle this problem in the framework of options [Sutton,Precup and Singh, 1999; Precup, 2000]. We derive policy gradient theorems for options and propose a new option-critic architecture capable of learning both the internal policies and the termination conditions of options, in tandem with the policy over options, and without the need to provide any additional rewards or subgoals. Experimental results in both discrete and continuous environments showcase the flexibility and efficiency of the framework.

The process for transferring knowledge of multiple reinforcement learning policies into a single multi-task policy via distillation technique is known as policy distillation. When policy distillation is under a deep reinforcement learning setting, due to the giant parameter size and the huge state space for each task domain, it requires extensive computational efforts to train the multi-task policy network. In this paper, we propose a new policy distillation architecture for deep reinforcement learning, where we assume that each task uses its task-specific high-level convolutional features as the inputs to the multi-task policy network. Furthermore, we propose a new sampling framework termed hierarchical prioritized experience replay to selectively choose experiences from the replay memories of each task domain to perform learning on the network. With the above two attempts, we aim to accelerate the learning of the multi-task policy network while guaranteeing a good performance. We use Atari 2600 games as testing environment to demonstrate the efficiency and effectiveness of our proposed solution for policy distillation