Optimal Neural Codes for Estimation and Control


May 4, 2015 - 1:30pm
NW 243
About the Speaker
Ron Meir (Department of EE, Technion, Israel)

Optimal neural codes have been widely studied in the field of computational neuroscience. Extending static formulations we provide closed form solutions for dynamic encoding settings, characterizing optimal encoding strategies. Motivated by recent results from motor control, we argue that encoding strategies should depend on the task for which they are being used. Given the different requirements of purely perceptual (e.g., object classification) and action-oriented (e.g., object manipulation) tasks, it is expected that sensory adaptation should differ within these domains. We show, based on principles of optimal estimation and control, that sensory adaptation for control differs from sensory adaptation for perception, even for simple setups. This implies, consistently with experiments, that when studying sensory adaptation, it is essential to account for the underlying task. In a broader setting, top-down effects, task-dependent and otherwise, have been shown experimentally to contribute to early sensory processing, thereby inspiring the development of a theoretical framework for such phenomena.