Develop information theoretic methods, based on the new concept of intersection information, to measure how much tactile information carried by a spiking neural code is used to implement specific behaviors in perceptual-guided tasks. These methods, which are unique in being able to determine the behavioral relevance of neural codes and not only their sensory information content, will be applied to both neural spike trains and artificial sensors. Building on recent technical advances, we will use the partial information decomposition framework to isolate the the part of the redundant information that stimulus and spiking response share about the appropriate choice that is also a part of stimulus-spikes information. These methods will be specifically adapted to address one of the main sensory coding problems of this project: how best to encode at the sensor/neuron population level, and then use for discrimination tasks, tactile information that is temporally precise at specific times but that needs to be integrated over long time scales for discrimination (e.g. long temporal sequences of tactile stimuli). Following recent progress on this issue on auditory and visual population coding of task relevant signals, we will explore the idea that codes exploiting either correlations across time or correlations across cells can generate long time scales of neural representations that make it easier to both encode past and present information in instantaneous activity, and to facilitate its behavioral readout. This will be achieved by using first machine learning methods to reduce the dimensionality of population time sequences of spikes coupled with correlation shuffling methods to assess population coding and its modulation by correlation.

Provide fundamental mathematical tools that can be used to determine the tactile population codes based on spike timing and used by biological systems for performing specific tasks, and that would be optimal in robots to encode task-relevant tactile information

University of Sheffield

S. Panzeri, R. Haschke, S. Micera, C. Bartolozzi

COMP PRO ROB TECH