PhD Students

PhD Project Suggestion: Computational Modelling of Cognition

Howard Bowman


Computational modelling techniques have made an important contribution to psychology and the brain sciences in general. The techniques that have been applied to such modelling can be divided into symbolic and sub-symbolic approaches. The former of these is largely dominated by the ``good old fashioned AI'' tradition, as exemplified by the unified theories - SOAR and EPIC, while the latter is dominated by neural network based techniques.

A central issue in this field is the choice of abstraction level. For example, there is important research on relating different levels of abstraction, e.g. relating symbolic and sub-symbolic models using hybrid approaches and developing lower level connectionist models, such as the anatomically prescribed neural network models of Arbib and Rolls.

As part of an ongoing collaboration between the Computing Laboratory and Psychologists at the Cognition and Brain Sciences Unit in Cambridge, Experimental Psychology at Cambridge and Experimental Psychology at Birkbeck College (University of London) computational modelling of attentional processes and visuo-motor control are being investigated using a number of different techniques. For example, process algebra and logic are being used to model the, so called, attentional blink, while neural network modelling of inhibitory processes is being investigated.

From this platform of existing research, there are a number of different directions for PhD research that could be undertaken. These include,

  1. Psychological level connectionist modelling of attention and perception. This would continue the thread of work currently being undertaken in collaboration with institutions in Cambridge and at Birkbeck College.
  2. Brain level connectionist modelling. There is some evidence that the Basal Ganglia are implicated in the inhibitory processes under investigation in some of our modelling work on motor control. A potentially very rewarding next step would be to build a neural network model of structures in the Basal Ganglia and try to relate this model to our existing psychological level model of inhibition.
  3. Relating abstraction levels. An important topic is how to relate computational models at different levels of abstraction, e.g. how to relate symbolic and sub-symbolic models. We believe that notions of refinement in process algebra and logic can be beneficially applied to this problem.