Taco Cohen

University of Amsterdam, Qualcomm (United Kingdom)

Papers

Total Citations

H-Index

About

Taco Cohen is a leading researcher at the intersection of geometric deep learning, equivariant neural networks, and representation learning, whose work has fundamentally shaped how machine learning models handle structured and spatially complex data. He is perhaps best known for his contributions to spherical convolutional neural networks, extending the power of CNNs to non-planar domains such as omnidirectional imagery used in robotics and drone navigation — work that has garnered 38 citations and become a foundational reference in geometric deep learning. Cohen's research consistently tackles the challenge of building neural architectures that respect the underlying geometry of data, as demonstrated by his Geometric Algebra Transformer, which offers a unified framework for handling diverse geometric quantities across physics, chemistry, and computer vision. His work on harmonic exponential families provides flexible probabilistic tools for data on manifolds, addressing a critical gap in fields like molecular biology and geoscience. More recently, Cohen has expanded into causal representation learning and robotic affordance discovery, exploring how agents can efficiently learn structured world models from limited supervision. Across these diverse contributions, his research reflects a coherent vision: grounding machine intelligence in the mathematical structures that govern the physical world.