Fall 2016

Seminars are held at 11:30AM in Cullimore Hall, Room 611, unless noted otherwise. For questions about the seminar schedule, please contact Casey Diekman.


Date: November 22, 2016

Speaker: Elad Schneidman
Weizmann Institute of Science & New York University,

University Profile

Title: "A Thesaurus for Neural Population Codes and a Neural Metric for Stimulus Space"

Abstract:

The combinatorial space of possible neural population activity patterns and neuronal noise imply that direct mapping of a “dictionary” for stimuli and neural responses is practically impossible. To accurately read information from novel population activity patterns, the brain must therefore learn the organization of this space and identify which patterns carry similar information. We present a novel approach to learning a thesaurus for the codebook of a neural population, where instead of using common but somewhat arbitrary assumptions about syntactic similarity of neuronal codewords, we infer their semantic similarity by the overlap of stimuli they encode. Applied to the vertebrate retina, learning such a thesaurus reveals that the retinal code is organized in clusters of semantically similar activity patterns that may differ considerably in their structure or syntax. We suggest the brain may use this structure, which is highly reminiscent of the design of noise-robust codes in electrical engineering, and show how it allows the accurate decoding of novel stimuli. We further extend these ideas to learn a metric for stimulus space, based on the similarity of neural responses they elicit. We show (again) that this assumption-free neural-based metric differs considerably from the commonly used measures of similarity of sensory stimuli, and discuss its implications for neuronal decoding in the brain and neuronal prostheses.