Multi-Agent Reinforcement Learning

Saptarashmi is a Ph.D. student of computer science at the University of Maryland, College Park. He is working on the project of Multi-Agent Reinforcement Learning (MARL) to solve interdiction games of social impact where the goal is to apply, accelerate and scale the latest generalizable state-of-the-art ideas of artificial intelligence and reinforcement learning to address challenging social problems. The idea of MARL is to model multiple human/autonomous agents interacting with each other fostering competition and cooperation as its learning evolves. MARL has multiple applications in solving game theory problems like discovering, analyzing and disrupting illicit trafficking networks (be it arms, humans, wildlife, rare plants and deforested trees, illegal drugs, counterfeit goods), equitable markets, fair tax policies, climate change and existential disasters like pandemics. The goal of interdiction games is to assign limited checkpoints by defenders (e.g., rangers in a wildlife reserve) to catch attackers (e.g. elephant poachers) from targeting secure assets (e.g. elephants). The challenge becomes that the multiple agents in an ever-changing environment can interact with multiple options where the goal is to find an optimal policy to help defenders catch attackers and protect green assets.