Research Interests

The brain is made of billions of neurons, which together form the world's most powerful information-processing machine. Despite decades of research, the fundamental principle by which these cells work together is still unknown.

Many theories for brain function have been proposed over the last century. But only in the last few years has it become possible to record simultaneously from large enough numbers of neurons to put these theories to the test experimentally. This is an unprecedented opportunity, but it opens up a new question: how do we go from the gigabytes of experimental data that we now have, to concise conclusions about the function of the brain?

The data processing methods traditionally used in neuroscience are not sophisticated enough to exploit this new flood of information. Fortunately, modern statistics and machine learning theory is making great strides in precisely the type of techniques needed to process these large multivariate databases. By applying these methods to neuronal data, we can now test long-standing hypotheses about brain function.

The Cell Assembly

The main focus of our research is an experimental search for "cell assemblies". Before describing what a cell assembly is, it will be useful to describe what it is not.

The brain is often thought of as a feed-forward system. In this scheme, sensory information is processed by successive levels of cortical “analyzers”, each of which transforms the results of previous levels, until sensory information is in a suitable form to guide the animal’s behavior. In support of this idea, the pattern of connections in the cortex does appear to respect a hierarchical organization, with the output of low-level areas corresponding to a single sensory modality being integrated into high-level multi-modal areas. Responses in higher-level sensory areas appear to have more complex responses to sensory stimuli, in agreement with increased abstraction as the hierarchy is traversed.

However, there are several levels at which this feed-forward picture is incomplete. At the "circuit diagram" level, there more connections projecting across and down the hierarchy, than there are feed-forward projections. What's more, if information were processed in a strictly feed-forward manner, one would expect a neuron to respond identically to repeated presentations of the same sensory stimulus. Although this is a fairly good approximation in primary sensory areas of cortex, in high-level structures responses are often more variable than expected from strict sensory control. Finally, although feed-forward processing can describe how an animal could perform simple stimulus-response behaviors, it cannot explain more complex "top-down" behaviors such as memory or thought.

An alternative point of view, put forward over 50 years ago by Canadian psychologist Donald Hebb, holds that recurrent and feedback connections play an essential role in brain function. The principal actor in this view is the “cell assembly”, an anatomically distributed subset of neurons, amongst which mutually excitatory connections have been strengthened by repeated co-activation, allowing the assembly to later maintain its activity through reverberation without direct sensory stimulation. This theory allows for sensory-response behavior, and also behavior resulting purely from internally generated cognitive activity, by the sequential activation of a series of assemblies, leading in turn to the production of motion.

In our research, we search for signatures of assembly activity in simultaneous recordings from multiple neurons, and aim to characterize the properties of assembly activity in ways not possible from theory alone.