Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster D in Poster Session D: Wednesday, August 5, 2:00 – 3:45 pm, Kimmel Center, Shorin & Rosenthal Rooms

Signal-to-Noise Ratio Predicts Neural Response Predictability with Limited Association with Tuning Properties in Mouse Visual Cortex

Weiwei Wang¹, Wu Li¹, Tianyi Qian²; ¹Beijing Normal University, ²Qiyuan Lab

Presenter: Weiwei Wang

DNNs predict visual responses neuron by neuron, yet prediction quality varies widely, and the factors behind this variability remain unclear. Total response power seems a natural candidate for explaining predictability, yet it conflates signal and noise. Without decomposition, their contributions remain indistinguishable. We decomposed neural responses from mouse visual cortex (Sensorium 2023; N = 78,853 neurons) into signal and noise, and related their ratio (SNR) to the predictive performance of the ViV1T model. SNR (Spearman ρ = 0.552) predicted performance more strongly than total response power (ρ = 0.310); signal power alone (ρ = 0.410) also exceeded the latter. Signal and noise power were strongly correlated (ρ = 0.931), so their ratio captured predictability better than either alone: high-noise neurons with strong signals remained predictable (r = 0.401). SNR showed limited association with classical tuning (OSI ρ = 0.093; DSI ρ = 0.045), reflecting an encoding property beyond selectivity. We propose SNR as a quantitative criterion for neuron selection in encoding models.

Topic Area: Methods, Tools, Theory & Neural Coding