Overview
- The Science Advances paper by Moosavi, Hindupur and Shimazaki published Sept. 25, 2026 reanalyzed recordings of roughly 18,000–21,000 neurons from mouse primary visual cortex to measure how linearly decodable information scales.
- The team repeatedly subsampled neural populations and used eigendecomposition to separate collective noise into activity modes and to track how the stimulus projects into those modes.
- They identified two scale‑invariant power laws that govern how noise strength and signal–noise alignment change with population size and that predict continued information growth even as noise slows accumulation.
- A general mathematical framework in the paper shows information scaling depends on the full noise eigenspectrum and proves that, under subsampling, genuine information‑limiting correlations would appear as differential correlations.
- The findings point to design principles for fault‑tolerant neuromorphic and distributed computing systems but rest on linear decoding, extrapolation from mouse V1 data, and a small number of animals so engineering and cross‑species translation remain preliminary.