Stefano Fusi

The complexity of real world behavior cannot be reduced to a simple mapping between a specific stimulus and a certain response. The same stimulus can lead to different behaviors depending on the situation, the intention, and the task to be performed. For example, depending on whether we are traveling by bike or by car, the same traffic sign can have two completely different meanings: in one case it might prescribe to stop, whereas in another it might indicate that the road is reserved for us (e.g. highways are reserved for cars, and pedestrian areas are accessible only to bikes). The same visual stimulus corresponds to two different behaviors (go or stop) depending on the context (driving a car or riding a bike). We keep constantly in mind what the context is and we act according to the corresponding rules. There is accumulating evidence that this active representation of the context resides in prefrontal cortex (PFC). PFC receives inputs from all sensory modalities and it is directly connected to the premotor cortex, the superior colliculus and the cerebellum, all areas which control the final motor response. Recent electrophysiological experiments indicate that neurons in PFC respond selectively not only to the visual stimuli and to the planned motor responses, but also to the abstract rules which guide behavior. In some experiments the subject is instructed to follow a specific rule, or the rule is specified by a cue. In others the visuo-motor associations are learned from experience (e.g. by means of reinforcement learning) and the rules emerge automatically from the repeated occurrence of several visuo-motor associations. The mechanisms leading to the formation of rule representations in PFC are still largely unknown. Their importance clearly goes beyond the encoding of abstract rules considered here. Indeed it is most likely that the same mechanisms underlie other high cognitive functions like memory- and task-guided attention: focusing on one stimulus or ignoring a feature which is irrelevant for the task can be regarded as an expression of specific rules. The aim of our group's research is to investigate the mechanisms responsible for encoding the rules with realistic, detailed models of networks of neurons connected by plastic synapses. The rules would emerge spontaneously from the temporal statistics of the visuo-motor associations and would determine a context dependent behavior. Heterogeneity allows to represent a large number of combinations of visuo-motor associations, which are the basic ingredients to create the representations of the context.

lab page / publications