Neural architectures: what are they good for anyway?
UCL NeuroAI seminar series
2025-02-12
2025-02-12
Abstract
I discuss how we can ask the question of what different neural architectures are good for, and why this is challenging from the point of view of causal reasoning. My answer is that theoretical approaches can achieve something here that purely experimental approaches cannot, which is to compare to counterfactual models. I illustrate by talking about our work on modularity, and multimodal architectures.
Slides
Links
Related people
Related publications
2026
-
Ghosh M, Goodman DFM (2026)
Partial recurrence enables robust and efficient computation.
Communications AI & Computing
2025
-
Anil S, Goodman DFM, Ghosh M (2025)
Fusing multisensory signals across channels and time.
PLoS Computational Biology -
Béna G, Goodman DFM (2025)
Dynamics of specialization in neural modules under resource constraints.
Nature Communications
2024
-
Ghosh M, Béna G, Bormuth V, Goodman DFM (2024)
Nonlinear fusion is optimal for a wide class of multisensory tasks.
PLoS Computational Biology
Categories