Computational Neuroscience

Selective attention is nature's answer to a computational dilemma. On the one hand, complex organisms need a multitude of sensors to provide information that at some time is important, even crucial. For instance, you really want to know if an insect with the potential of injecting poison into you crawls over the back of your hand. On the other hand, most of the time you couldn't care less what the back of your hand feels like and processing in detail all information available from the sensors at all times would require vast computational resources. (Your brain is an evolutionarily expensive organ, it consumes on the order of ¼ of all the energy that your body provides, and you don't want to grow it any bigger than necessary!).

The answer to this dilemma is to have lots of sensors but not to use them most of the time! In other words, while information from the peripheral nervous system is collected continuously, only a very small part of that eventually influences behavior. While you are sitting at your desk and reading this text, the information sent to the brain by the sensors on the back of your hand is discarded, except for the very rare times when it is actually useful (like when the bee is crawling over it). Of course, this is a very simple example but this is the main idea: we need a filter to select the useful information out of the deluge of available information sent in to the brain at every second by the sensors. This filtering process is called selective attention. Generations of psychologists have studied this selection process in much detail and have shown that it consists of an interplay of 'bottom up' processes, in which the presented stimulus determines where the person or animal attends, and 'top-down' processes in which the focus of attention is determined by volition.

Several problems have to be solved. Perhaps the most obvious one is to decide which information is to be retained and which is to be discarded. We are in the process of developing detailed computational models of how this happens in the visual modality where more is known than in any other. Although this is very difficult to do in complete generality, we have shown that one can make surprising progress even with very simple assumptions based on bottom-up processes

A different problem is to answer the question, how this information is actually represented in the brain. Let us consider two situations in which exactly the same stimulus is being presented, but in one case it is attended and in the other it is unattended. Psychologists have shown that the behavior of a person is different in these two cases; for instance, if the stimulus presented is a light and the subject is instructed to push a button when the light starts to dim, the subject will react faster if the stimulus is attended than when it is unattended. If there is a difference in the behavior, there must be a difference in the neural representation of the stimulus! We are studying what this difference consists in. For a variety of reasons, we believe that this difference is in the synchrony structure of the neural spike trains that the stimulus gives rise to.

We have also started to study human eye movements which are a different form of attention (usually you attend to where you look and vice-versa). It turns out that the experiments we undertook also lead to practical applications in video compression, telemedicine etc.

There are more fields we are working on and you are invited to browse the lab homepage.