Physics (Internal)

Models of the insect brain for odor discrimination and decision making

This event is part of the Biophysics/Condensed Matter Seminar Series.

Abstract:

In the course of evolution animals, bacteria and plants have developed
sophisticated methods and algorithms for solving difficult problems in
chemical sensing very efficiently. Complex signalling pathways inside
single cells can trigger movement toward the source of a nutrient.
Complex networks of neurons are able to compute odor types and the
distance to a source in turbulent flows.  These networks of neurons use
a combination of temporal coding, layered structures, simple Hebbian
learning rules, reinforcement learning and inhibition to quickly
learn about chemical stimuli that are critical for their survival.

In this talk we revisit the critical elements of the insect brain involved
in odor discrimination and determine the impact that each of the areas have
in learning an odor discrimination task. We apply these lessons to the
problem of gas identification with artificial sensor arrays.
The insect brain must cope with those conditions by preprocessing
the data using the excitatory-inhibitory network in the first
relay station of the insect olfactory system.It manages to
extract and dynamically inhibit common odor representation and
enhances the sensitivity to novel ones. Thus, we use the insect
olfactory system as a base and inspiration to build odor recognition
devices that take advantage of the spatio-temporal characteristics of
the turbulent gas plume. We demostrate how they can be merged and
compared to state-of-the-art machine learning algorithms.