Key points are not available for this paper at this time.
We formulate a basic problem for neural encoding: the stimulus should be accurately represented in the neural responses. We use this criterion to design the optimal receptive fields of a model visual system. Since reconstruction fidelity is an ensemble average over signals and noise, the statistics of natural stimuli play a central role. We compare our results with those of similar studies which apply optimization principles based on information theory.
Daniel Ruderman (Sat,) studied this question.