Selected Publications

The Price of Paranoia: Robust Risk-Sensitive Cooperation in Non-Stationary Multi-Agent Reinforcement Learning

Deep Kumar Ganguly, Chandradithya S Jonnalagadda, Pratham Chintamani, Adithya Ananth Accepted at the Adaptive and Learning Agents (ALA) Workshop, AAMAS 2026 Preprint

Abstract

We introduce RATTL (Robust Adaptive Trust-Region Learning), an algorithm designed to sustain cooperation in non-stationary MARL by modulating updates based on partner unpredictability. By resolving the EVaR Paradox and targeting policy gradient update variance, we provably expand the cooperation basin in coordination games.

BibTeX

Show BibTeX

@misc{ganguly2026priceparanoiarobustrisksensitive,
      title={The Price of Paranoia: Robust Risk-Sensitive Cooperation in Non-Stationary Multi-Agent Reinforcement Learning}, 
      author={Deep Kumar Ganguly and Chandradithya S Jonnalagadda and Pratham Chintamani and Adithya Ananth},
      year={2026},
      eprint={2604.15695},
      archivePrefix={arXiv},
      primaryClass={cs.GT},
      url={[https://arxiv.org/abs/2604.15695](https://arxiv.org/abs/2604.15695)}, 
}