A Deep Q-Learning Agent for the L-Game with Variable Batch Training

TL;DR

This paper presents a Deep Q-Learning approach with variable batch training to effectively train an agent for the L-Game, overcoming challenges of sparse rewards and large action spaces without domain knowledge.

Contribution

It introduces variable batch size training in Deep Q-Learning to improve learning efficiency in environments with sparse rewards and large action spaces.

Findings

Successful training of a strong L-Game agent without search methods

Variable batch training accelerates learning process

Effective handling of sparse rewards in low-dimensional state spaces

Abstract

We employ the Deep Q-Learning algorithm with Experience Replay to train an agent capable of achieving a high-level of play in the L-Game while self-learning from low-dimensional states. We also employ variable batch size for training in order to mitigate the loss of the rare reward signal and significantly accelerate training. Despite the large action space due to the number of possible moves, the low-dimensional state space and the rarity of rewards, which only come at the end of a game, DQL is successful in training an agent capable of strong play without the use of any search methods or domain knowledge.