Multiagent Approaches to Enhance Learning and Trust in AI Systems

This thesis addresses two central challenges in artificial intelligence: achieving scalable learning in interactive environments and ensuring trust in real-world deployment. Many domains involve multiple adaptive agents interacting under uncertainty and limited information. To be effective, agents must adapt continuously while operating efficiently in large and complex decision spaces. Using frameworks such as extensive form games and multiagent reinforcement learning, this work demonstrates how regret minimization can serve as a unifying foundation for scalable learning. Contributions include a new interpretation of an existing Multiagent Reinforcement Learning (MARL) as a regret-based method, improvements its learning through modified update rules, and the introduction of EINR, an algorithm that applies multiplicative weights in extensive form games to achieve faster and more reliable convergence. The thesis also develops methods for trustworthy AI, where robustness, interpretability, and fairness are as important as accuracy. Building on cooperative game theory and social choice theory, it introduces Banzhaf indices to generate stable counterfactual explanations for graph neural networks and proposes voting based aggregation rules such as thresholded Borda count to defend against data poisoning in ensemble learning. Taken together, these contributions show how multiagent approaches can enhance both learning and trust, offering a pathway to AI systems that are capable, transparent, and reliable in practice.

Author dc:creator

• Chirag Chhablani (23291350)

• Computer Science

Statement dc:rights

• In Copyright

DOI dc:identifier

https://doi.org/10.25417/uic.31451101.v1

OAI identifier oai:identifier

oai:figshare.com:article/31451101