Matthew W. Hoffman

University of Cambridge, University of Washington

Papers

Total Citations

500

H-Index

About

Matthew W. Hoffman is a leading researcher in Bayesian optimization and multi-task reinforcement learning. His most influential contribution is the development of **Predictive Entropy Search (PES)** , a groundbreaking information-theoretic framework for efficiently optimizing expensive black-box functions. This work, which has garnered over **400 citations**, fundamentally advanced the field by enabling practitioners to select evaluation points that maximize information gain about the global optimum, solving a previously intractable acquisition function problem. Beyond optimization, Hoffman has made significant strides in artificial intelligence, notably with the **Intentional Unintentional (IU) agent** (2017), which extended deep deterministic policy gradients to solve multiple continuous control tasks simultaneously—a core challenge in general-purpose AI. His earlier research on **gaze imitation and shared attention** (2006) explored probabilistic models of social learning, demonstrating a sustained interest in how agents learn from their environment and from others. Hoffman’s work sits at the intersection of probabilistic modeling, decision theory, and autonomous learning, providing both foundational theory and practical algorithms that continue to shape modern machine learning research.