Reproduce “Explanatory models in neuroscience: Part 2 -- constraint-based intelligibility” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Rosa Cao, Daniel Yamins
This paper advances a framework for "constraint-based intelligibility" in neuroscience, demonstrating how computational models like goal-optimized neural networks can explain *why* brain systems are structured as they are by virtue of the challenging behavioral tasks they are optimized to perform, accurately predicting neural activity without direct neural data training.
Key insights
- Intelligibility in neuroscience can be achieved by understanding how top-down constraints (e.g., behavioral goals) shape biological systems, complementing bottom-up mechanistic explanations.
- The "Contravariance Principle" suggests that harder computational goals impose stronger constraints, leading to a smaller, more convergent set of solutions, thus making model-to-brain matching easier.
- Optimization processes in artificial neural networks can serve as a powerful proxy for biological evolution and development, explaining why independent systems converge on similar solutions for similar problems.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.