A Framework of Mapping Visual Stimulus to Neural Responses via Discrete Latent Representation
A V1 model that passes images through a small codebook of discrete features predicts neurons’ responses as well as a standard CNN, as long as the readout is factorized into where and what.