Deep Abstract Q-Networks
This addresses a challenging problem in reinforcement learning for domains like video games, but it is incremental as it builds on existing abstraction methods.
The paper tackles the problem of learning and planning in high-dimensional domains with long horizons and sparse rewards, such as Montezuma's Revenge and Venture, by combining deep reinforcement learning with expert-provided state abstraction. The result is that their algorithm significantly outperforms previous methods like Deep Q-Networks on these domains, exhibiting backtracking behavior not seen before.
We examine the problem of learning and planning on high-dimensional domains with long horizons and sparse rewards. Recent approaches have shown great successes in many Atari 2600 domains. However, domains with long horizons and sparse rewards, such as Montezuma's Revenge and Venture, remain challenging for existing methods. Methods using abstraction (Dietterich 2000; Sutton, Precup, and Singh 1999) have shown to be useful in tackling long-horizon problems. We combine recent techniques of deep reinforcement learning with existing model-based approaches using an expert-provided state abstraction. We construct toy domains that elucidate the problem of long horizons, sparse rewards and high-dimensional inputs, and show that our algorithm significantly outperforms previous methods on these domains. Our abstraction-based approach outperforms Deep Q-Networks (Mnih et al. 2015) on Montezuma's Revenge and Venture, and exhibits backtracking behavior that is absent from previous methods.