Graying The Black Box Understanding Dqns

# Graying the black box: Understanding DQNs

**arXiv ID:** 1602.02658
**Authors:** Tom Zahavy, Nir Ben Zrihem, Shie Mannor
**Published:** 2016-02-08T17:27:31Z
**Abstract:**
In recent years there is a growing interest in using deep representations for reinforcement learning. In this paper, we present a methodology and tools to analyze Deep Q-networks (DQNs) in a non-blind matter. Moreover, we propose a new model, the Semi Aggregated Markov Decision Process (SAMDP), and an algorithm that learns it automatically. The SAMDP model allows us to identify spatio-temporal abstractions directly from features and may be used as a sub-goal detector in future work. Using our tools we reveal that the features learned by DQNs aggregate the state space in a hierarchical fashion, explaining its success. Moreover, we are able to understand and describe the policies learned by DQNs for three different Atari2600 games and suggest ways to interpret, debug and optimize deep neural networks in reinforcement learning.

## Skill Description

This skill is generated from the arXiv paper: Graying the black box: Understanding DQNs (1602.02658).

## How to Use

[To be filled in by the user or by future automation]

## References

- [arXiv:1602.02658](http://arxiv.org/abs/1602.02658v4)