CLAM model of=
visual
search
Background
Visual search is an important topic in psychology and neuroscience. It is a window on how humans process complex visual scenes. In visual search visual sensory information competes with internal goals and p= rior information. The sensory information shows us when something new and possib= ly important is present. It may require immediate action. Internal goals and p= rior information help to actively process visual information: we only consider p= art of the visual sensory information that we receive: that part that we believ= e is relevant for what we want to do. The mechanisms by which we select only that part of the visual information that we believe is relevant allows us to ign= ore most information that enters our retina, so that we can concentrate on part= s of the visual input that really matter. But clearly there is some tension here= :
How do we know whether we can ignore novel sensory information? It may be about a lion about to eat us and although are goal m= ay be going to the cinema, we may have to discard i= t for a moment and deal with our sensory input. Clearly humans are very efficient= in dealing with the visual world, not only in processing sensory information, but also= in representing the world around us. We would like to be able to endow robots = an machines with some of this efficiency. In order to = be able to this, we must first understand how it works in humans.
As even visual search is quite a complex topic, this research area is broken down in several topics:
Project 1: A =
model of
‘pop-out’
Level: UG, MSC
Requirements: some exposure to Python programming= . Not be put off by programming.
Project 2: At=
tention
in an artificial model of biological vision
Level: MSc, PhD
Project 3: Ne=
ural
dynamics: the behaviour of large groups of spiking neurons
3a: Matching =
the
response curve of a simulated population of spiking neurons with population
density predictions
Level: UG, MSc
3b: Modelling
synaptic kinetics in population density techniques
Level: MSc, PhD
Requirements: Programming experience, willingness to l= earn C++ and Python and an interest in scientific computing. For this topic it is necessary to have a good mathematical background. Experience with MATLAB is= an advantage.
Project 4: Ne=
ural
representation of compositional relations
Project 4a: U=
nsupervised
learning of structural relations
Level MSc, PhD
Project 4b: T=
he use
of compositional representations in the learning of novel compositional obj=
ects
Level: UG, MSc