Neural mechanisms for resilient sensory computation under metabolic stress

Food-restricted mice can solve visual discrimination tasks despite reduced synaptic currents and increased noise resulting from cellular changes in the primary visual cortex. Combining neural recordings and modeling, we study the population structure of the noise and show it is mainly concentrated along task-irrelevant directions in high-dimensional population activity, hence preserving coding on the low-dimensional stimulus manifold. In addition, we predict that food restriction shortens decision times, conserving energy but limiting the evidence accumulation needed for fine discrimination.