Vivek Veeriah

Papers

Total Citations

H-Index

About

Vivek Veeriah is a researcher specializing in reinforcement learning, meta-learning, and representation learning, with a particular focus on developing adaptive algorithms that improve the efficiency and autonomy of learning agents. His work sits at the intersection of temporal-difference learning and stochastic optimization, addressing one of the field's most persistent practical challenges: how to automatically tune learning parameters without human intervention. Veeriah's most notable contributions include his investigations into step-size adaptation for temporal-difference (TD) learning. His papers on TIDBD and feature relevance learning through stochastic meta-descent propose principled methods for automatically adapting step-size parameters — hyperparameters that critically govern an agent's learning performance yet have historically required laborious manual tuning. These works, accumulating citations across the reinforcement learning community, demonstrate both theoretical rigor and practical utility. His doctoral thesis, *Discovery in Reinforcement Learning* (2022), represents a broader synthesis of his research agenda, examining how agents can autonomously discover useful knowledge structures within complex environments — a capability increasingly vital as RL is applied to demanding domains like robotics and game-playing. Veeriah's contributions reflect a commitment to building more self-sufficient, adaptable learning systems, positioning him as a thoughtful contributor to the foundations of modern reinforcement learning research.