Width-based planning methods have been shown to yield state-of-the-art performance in the Atari 2600 domain using pixel input. One successful approach, RolloutIW, represents states with B-PROST boolean feature set. An augmented version of pi-IW, shows that learned features can be competitive handcrafted ones for width-based search. In this paper, we leverage variational autoencoders (VAEs) lear...

The combination of modern Reinforcement Learning and Deep Learning approaches holds the promise of making significant progress on challenging applications requiring both rich perception and policy-selection. The Arcade Learning Environment (ALE) provides a set of Atari games that represent a useful benchmark set of such applications. A recent breakthrough in combining model-free reinforcement l...

We explore the use of Evolution Strategies, a class of black box optimization algorithms, as an alternative to popular RL techniques such as Q-learning and Policy Gradients. Experiments on MuJoCo and Atari show that ES is a viable solution strategy that scales extremely well with the number of CPUs available: By using hundreds to thousands of parallel workers, ES can solve 3D humanoid walking i...

Spiking neural networks (SNNs) have great potential for energy-efficient implementation of Deep Neural Networks (DNNs) on dedicated neuromorphic hardware. Recent studies demonstrated competitive performance SNNs compared with DNNs image classification tasks, including CIFAR-10 and ImageNet data. The present work focuses using in combination deep reinforcement learning ATARI games, which involve...

This paper presents the selective use of eye-gaze information in learning human actions Atari games. Extensive evidence suggests that our eye movements convey a wealth about direction attention and mental states encode necessary to complete task. Based on this evidence, we hypothesize eye-gaze, as clue for direction, will enhance from demonstration. For purpose, propose augmentation (SEA) netwo...

Machine learning algorithms for controlling devices will need to learn very quickly, with very few trials. Such a goal can be attained with concepts borrowed from continental philosophy and formalized using tools from the mathematical theory of categories. Illustrations of this approach are presented on a cyberphysical system: the slot car game, and also on Atari 2600 games.

We show how an ensemble ofQ-functions can be leveraged for more effective exploration in deep reinforcement learning. We build on well established algorithms from the bandit setting, and adapt them to the Q-learning setting. We propose an exploration strategy based on upper-confidence bounds (UCB). Our experiments show significant gains on the Atari benchmark.