Picture
Profile picture for user DENEVE S

Sophie Denève

LNC2

Faculty

29 rue d'Ulm
75005 Paris France

Laboratory
LNC2
GNT
Team
Normative approaches to neural circuits and behavior
Office
2nd floor, GNT office
Tel
+33 (0)1 44 32 26 35
Research Interests

Cortical neural responses are hugely variable, redundant, and robust. Does the brain relies on huge numbers of unreliable units to compensate for such "noise"? Are spikes random samples from an underlying firing rate? Our recent work, combining theory and analysis of multi-unit recordings in retina and cortex, suggests quite the contrary. Despite apparences, biological neural networks are exquisitely tuned to represent relevant stimuli as reliably and efficiently as possible, spike after spike.  The signature of this efficiency is the balance between excitation and inhibition maintained at all levels of cortical neural processing. This introduces an entirely new theoretical and experimental framework to explore neural coding, plasticity and adaptation.

Recent years have seen the growing use of models formalizing sensory perception, motor control or behavioral strategies as probabilistic inference tasks. Excitable neural structures face similar problems than behaving organisms: they receive noisy and ambiguous inputs, must accumulate evidence over time, combine unreliable cues, and compete with other neurons representing alternative interpretations of the sensory input. We apply such normative models, particularly Bayesian networks, in order to further our understanding of the function and dynamics of biological neural networks.

The E/I balance is equally important at a much larger scale, when we consider how different brain areas communicate. Cognition, motor action and decision making are inherently hierarchical, and this hierarchy is reflected in the brain organisation. We consider how E/I balance in a hierachical neural networks would affect perception, motor control and decision making at the macroscopic, behavioral scale. We find that it can cause circular inference, where prior beliefs are "mixed-up" and mistaken as sensory information, and vice-versa. We hypothesize that such a dis-regulation could be involved in the formation of hallucinations and delusions. We are currently testing this hypothesis by performing behavioral experiments in schizophenic patients and controls.   

Current Projects
  • Neural networks implementing hierarchical probabilistic models
  • Decision making with unknown sensory reliability
  • Are single neurons Bayesian integrators?
  • Efficient coding in balanced spiking networks
  • Computational Neuroscience of hallucinations