Carlos Diuk

About

Carlos Diuk is a leading researcher in artificial intelligence, with a primary focus on reinforcement learning, structure learning, and feature selection. His most influential work, "The Adaptive k-Meteorologists Problem and Its Application to Structure Learning and Feature Selection in Reinforcement Learning" (2009, 80 citations), makes three pivotal contributions. First, Diuk formalizes a novel problem within the KWIK (Knows What It Knows) learning framework, introducing the Adaptive k-Meteorologists Algorithm. He rigorously analyzes its sample-complexity upper bound and provides a matching lower bound, establishing theoretical foundations for efficient probabilistic concept learning. Second, he demonstrates how this algorithm can be applied to structure learning in reinforcement learning, enabling agents to discover underlying dependencies in complex environments. Third, Diuk shows its utility for feature selection, allowing agents to identify relevant state variables and ignore irrelevant ones. This work bridges theoretical guarantees with practical algorithmic design, offering a principled approach to scaling reinforcement learning to high-dimensional problems. His contributions continue to influence researchers working on sample-efficient learning and autonomous decision-making systems.

Research Focus

Key Achievements

Top Papers

1. 1
   The adaptive <i>k</i> -meteorologists problem and its application to structure learning and feature selection in reinforcement learning

   80 citations · 2009

Key Collaborators

Contact & Links