Ridge regression beats RSA as model-brain similarity metric across visual tests
dyamins · x · 2026-08-27
The author's Substack post reports detailed empirical comparisons of RSA vs ridge regression across a wide range of visual system tests, testing predictions of RSA contravariance theory:
- RSA and ridge regression correlate well overall, especially in high-quality fMRI data;
- The theoretically predicted Ridge-RSA formula works decently;
- But strong behavioral tests expose issues with raw RSA: ridge-based model-brain similarity is uniformly more predictive of behavioral match across many metrics;
- The same holds for low-level vision: ridge predicts 20+ V1 physiology metrics better than RSA.
The author argues the field is finally building a clearer picture of how these similarity measures relate.
More from Research
- Sai Agent Tops OSWorld 2.0 Benchmark at Lower Cost — TianbaoX · 2026-08-28
- Diverse pretraining reduces need for embodiment-specific data in robotics — chris_j_paxton · 2026-08-28
- PACT benchmark: one sentence of pressure raises AI rule violations 65%; no model clears unsupervised bar — baseten · 2026-08-28
- Two New Benchmarks Open-Sourced for Testing Agents in Dynamic Environments — AIwithGhotai · 2026-08-28
- Alex Rives, pioneer of protein language model ESM, named to TIME100 AI — proteinrosh · 2026-08-28
- New Paper: Dynamic Multi-Byte Prediction Speeds Up Hierarchical Byte-Level LMs — madeofAjala · 2026-08-28