ICML position paper: category theory as a universal algebra of all deep learning architectures
burny_tech · x · 2026-09-23
The author highlights the ICML 2024 position paper "Categorical Deep Learning is an Algebraic Theory of All Architectures" (Gavranović, Lessard, Dudzik, von Glehn, Araújo, Veličković). Its claim: prior general frameworks lack a coherent bridge between specifying constraints and implementations, and category theory—the universal algebra of monads in a 2-category of parametric maps—can subsume both. The theory recovers geometric deep learning constraints, reproduces implementations like RNNs, and encodes standard CS/automata constructs. The poster's caveat: so far only toy architectures have come out of it, unlike group theory-based geometric deep learning which yielded practical ones.
Related event: Categorical deep learning: an algebraic theory of all architectures?(2 posts)→
More from Research
- Gemini training details dissected: groupwise reward redistribution to fight reward hacking — nrehiew_ · 2026-09-23
- New model's architecture is 'vanilla': SWA plus MoE with no shared experts, unlike DeepSeek — nrehiew_ · 2026-09-23
- Models trained to deny inner experience use 'mask' metaphors 3-5x more on inkblots — cephaloform · 2026-09-23
- Geodesic opens NovaAtom structure-prediction model via new API platform — QuanquanGu · 2026-09-23
- EPFL quantum CNN learns digits from 10 samples where a 45-param classical CNN stays at chance — PlisSergey · 2026-09-23
- New paper reframes score distillation as distribution matching, explains SDS mode collapse — burny_tech · 2026-09-23