A Social Neuro-AI approach to Multi-Agent Reinforcement Learning

Zahra Sheikhbahaee, Juan David Vargas Mazuera, Samuele Bolotta, Adam Safron, Leonardo Christov-Moore, Nicco Reggente, Dianbo Liu, Guillaume Dumas

Abstract: Inspired by cognitive science, we integrate a causal modular architecture in a multi-agent reinforce- ment learning (MARL) featuring specific constraints (i.e. recurrent neural networks), representing knowledge systems that can learn environment dynamics effectively.