Mark Van Rossum
Wednesday 4th September 2013
Time: 4pm
Basement Seminar Room
Alexandra House, 17 Queen Square, London, WC1N 3AR
Plasticity and homeostasis of intrinsic excitability
In models of neural computation the relation between the neuron output
(the firing rate) and the net input is usually fixed. However, the
excitability is known to be modulated on short and long time-scale both
in 'Hebbian' as well as regulatory, homeostatic fashion. Here we study
both effects. First, in line with experimental data, we study how short
time changes in intrinsic excitability can assist learning of delay
conditioning tasks. Enhanced excitability can act to tag recently
active cells. Secondly, we study homeostasis of intrinsic excitability.
Homeostasis has often been introduced alongside plasticity to ensure
stability of neuronal networks and maximize information processing
capability. We introduce a framework for linear
homeostatic controllers. We show that stability of a single neuron does
not guarantee stability of a network of neurons. In particular, we find
that slow oscillations can develop that completely defeat the purpose of
homeostasis. Moreover we find that, in an effort to model the biological
cascade more closely, adding filters to the feedback can also have
de-stabilizing effects on the network level. These results constrain
biological and engineered homeostatic controllers.
Short CV:
Mark van Rossum received in PhD in statistical physics from the
University of Amsterdam. After postdocs at UPenn with Robert Smith and
Brandeis with Gina Turrigiano, he came to the University of Edinburgh in
2002. His research interests are sensory coding, synaptic plasticity,
data analysis, and the role of noise in the nervous system.