What shapes the discrete domains? Surround suppression in orientation space! Neurons preferring orientations flanking task stimuli are less responsive, creating a W-shaped modulation reshaping representations. Distortions not present in Naïve 🐭, their orientation representation space was more “flat”

Does activity in these domains drive behavior? We computed a "Go evidence" ratio: relative activation of Go & NoGo domains. It explained ~90% of the variance of perceptual decisions, as if V1 outputs probabilistic Go/NoGo estimates! Suggests discrete, domain-based integration of V1►sensory decisions

When task is difficult (NoGo° close to Go°), V1 doesn't encode stim orientation faithfully. Discrete domains emerge: "Go" (~45°) & a "NoGo" (~90°) in which the activity is constrained. Surprisingly NoGo resp. are bimodal-hinting at categorical processing. Not observed in Naïve 🐭 passively viewing

Behavioral thresholds vs. neuronal capacity: Mice trained in a Go/NoGo orientation task showed discrimination threshold between 15°–30°. But V1's neural activity contained more information; shallow neural nets trained on V1 data vastly outperformed behavior
➡️ at 20° angle, SNN D'>4, Mouse D'< 1.7

Through task training, V1 transitions from continuous to categorical representations that predict the 🐭 behavioral performance. Task-driven excitability modulation splits the evoked activity into discrete domains, suggests a probabilistic encoding of the stimuli category (Go/NoGo). More details in 🧵

Am I too late to the party?