MPI-INF Logo
Campus Event Calendar

Event Entry

What and Who

Updating Artificial Neural Networks: Translating Recent Discoveries about the Electrophysiology of Neurons into the Language of Computation

Alan Schoen
International Max Planck Research School for Computer Science - IMPRS
PhD Application Talk
AG 1, AG 2, AG 3, AG 4, AG 5, SWS, RG1, MMCI  
Public Audience
English

Date, Time and Location

Monday, 26 October 2015
09:00
120 Minutes
E1 4
024
Saarbrücken

Abstract

Artificial neural networks (ANNs) are widely used in machine learning, but they are only partially connected to biological research in neuroscience. In some cases, biological research has advanced the state of computer science. One example of this is deep learning, which was inspired by the multi-layer structure of visual cortex. On the other hand, biological research about computation within individual neurons has advanced in recent decades, but it has not transferred to ANNs. The structure of each individual neuron in an ANN is still based on a biological theory that is over 70 years old.

New experimental methods like patch-clamps, optogenetics, and two-photon imaging reveal new behaviors and computational abilities. In particular, we see that dendrites can organize and filter incoming signals in ways that greatly increase the complexity of each neuron. These findings could advance computer science, but they remain silioed in biology. I believe this is because they are expressed in a language of biology and electrophysiology that is inaccessible to computer scientists.
The goal of my masters thesis research was to translate recent biological findings into the language of computation. We used patch clamps to inject time-varying currents into neurons and study their response. Then I developed a theoretical computational framework to account for these findings which was consistent with findings from other experimental methods. The result was a mixed linear/nonlinear framework which simplified our understanding of the data.  I also proposed a computational structure that could explain the function of some neurons that process time-signals. I believe that more research in this area can strengthen the connection between biological neuroscience and theoretical computational neuroscience, to the benefit of both fields.

Contact

Andrea Ruffing
--email hidden
passcode not visible
logged in users only

Andrea Ruffing, 10/23/2015 18:51 -- Created document.