New metric 'Geometric Stability' reveals how brain regions encode reliable stimulus representations
A new study shows geometric stability predicts behavior better than temporal drift in neural codes.
A new preprint by Prashant C. Raju (arXiv, June 2026) introduces “geometric stability” as a distinct metric for neural population codes, separate from traditional temporal stability or decoding accuracy. Instead of asking whether population centroids drift over time, geometric stability measures how reliably the pairwise distance structure among stimuli reproduces across independent observations within a single session. The author formalizes this using the Spearman rank correlation between split-half representational dissimilarity matrices (RDMs). Across 229 area-session observations spanning 68 brain regions from the Steinmetz et al. 2019 visual discrimination dataset, geometric stability predicted trial-by-trial neural-behavioral coupling significantly (ρ = 0.18, p = 0.005), while centroid drift showed no effect (ρ = 0.002, p = 0.976). The regional hierarchy ran opposite to temporal stability: the striatum was most stable (mean S = 0.44), the hippocampus least (0.19).
Directionally consistent results from olfactory data (Bolding & Franks 2018) motivated an attractor network model. In the model, recurrent excitatory coupling amplifies split-half RDM consistency by completing stimulus patterns from sparse feedforward input—achieving a strong correlation of ρ = 0.64 (p = 0.010). This provides a circuit-level explanation for how geometric stability emerges and demonstrates its orthogonality to temporal drift. Raju argues that geometric stability offers a functionally relevant, circuit-dependent property of neural population codes, complementing recent work on how recurrent connectivity balances stability with sequential dynamics in hippocampal circuits. The finding has implications for designing brain-machine interfaces and understanding representational reliability in cognitive disorders.
- Geometric stability, measured as Spearman rank correlation between split-half RDMs, independently predicts neural-behavioral coupling (ρ=0.18, p=0.005) where centroid drift fails (ρ=0.002).
- Regional hierarchy: striatum most stable (S=0.44), hippocampus least (S=0.19), running opposite to temporal stability hierarchy across 68 brain regions.
- Attractor network model shows recurrent excitatory coupling drives split-half RDM consistency (ρ=0.64, p=0.010), providing a neural circuit mechanism.
Why It Matters
Geometric stability offers a more behaviorally relevant metric for neural code reliability, with potential applications in neural prosthetics and understanding cognitive disorders.