Research
The projects have ranged from robots and reinforcement learning to causality, category theory, and computational creativity. Across them runs one persistent question: what structure must an intelligent system construct in order to understand, decide, and revise?
Reinforcement learning, hierarchical control, multi-agent systems, robot learning, and optimization for agents that must act under uncertainty.
This work asks how decision problems can expose reusable structure rather than remain a collection of isolated tasks.
Proto-value functions, spectral methods, manifold alignment, transfer learning, and the geometry of high-dimensional data.
The central premise is that useful coordinates can be discovered from the topology and dynamics of a problem, rather than supplied in advance.
Causal inference from language, geometric causal discovery, trustworthy foundation models, and models that support intervention as well as observation.
The goal is to move from pattern extraction toward systems whose claims, mechanisms, and transformations can be inspected.
Diagrammatic backpropagation, LINCS, infinitesimal causality, double-involution learning, and controlled theory extension.
This program treats composition as a first-class obligation and asks how learning systems can diagnose, repair, and sometimes enlarge their own structural theories.
Current research program
The categorical-AI trilogy is both a synthesis of earlier work and a forward-looking laboratory for compositional machine learning, causal reasoning, and scientific creativity.
Explore the digital editions ↗