Towards sample-efficient episodic control with DAC-ML

TL;DR

This paper introduces DAC-ML, a new cognitive architecture inspired by DAC theory, incorporating hippocampus-like memory to improve sample efficiency in reinforcement learning tasks.

Contribution

The paper presents a novel architecture that leverages hippocampus-inspired memory for rapid policy learning, addressing sample inefficiency in reinforcement learning.

Findings

DAC-ML rapidly converges to effective policies in a foraging task.

Incorporating hippocampus-inspired memory improves learning speed.

Outperforms existing methods in sample efficiency.

Abstract

The sample-inefficiency problem in Artificial Intelligence refers to the inability of current Deep Reinforcement Learning models to optimize action policies within a small number of episodes. Recent studies have tried to overcome this limitation by adding memory systems and architectural biases to improve learning speed, such as in Episodic Reinforcement Learning. However, despite achieving incremental improvements, their performance is still not comparable to how humans learn behavioral policies. In this paper, we capitalize on the design principles of the Distributed Adaptive Control (DAC) theory of mind and brain to build a novel cognitive architecture (DAC-ML) that, by incorporating a hippocampus-inspired sequential memory system, can rapidly converge to effective action policies that maximize reward acquisition in a challenging foraging task.