Module details
CO636 Cognitive Neural Networks (15 credits)
Syllabus
- Introduction to cognitive neural networks.
The basic
motivation for cognitive neural networks. Neural networks will be placed
into a historical perspective related to symbolic approaches and in the
context of the artificial intelligence hypothesis. An overview of the
general approach to be taken throughout the course will be given. The course
text O'Reilly and Munakata "Computational Explorations in Cognitive
Neuroscience" will be introduced.
Practicals: Students will familiarise
themselves with the Leabra environment.
- The individual neuron.
Developing the idea of the
components of a neuron as a "detector". Neural networks will be explained in
terms of the biology of the brain at a cellular electro-transmission level.
This will be followed by abstracting the neurobiology into an initial neural
network framework, i.e. a set of mathematical equations.
Practicals:
Students will run single neuron simulations and appraise their level of
understanding. Students will work through the exercises/explorations in
Chapter 2.
- Networks of neurons.
A general framework for neural
network architectures both at an abstract level and in terms of networks in
the cortex. Unidirectional (feedforward) and bi-directional (recurrent)
interactions will be explained together with inhibitory mechanisms.
Practicals: Students will work through the explorations in Chapter 3.
- Model Learning.
These lectures will provide the
theoretical outline of a simple Hebbian model of learning, pertaining to
neurobiology and neural networks. It will also introduce other models of
unsupervised learning.
Practicals: Students will work through the
explorations in Chapter 4.
- Task Learning.
Error-driven task learning; the delta
rule and back propagation will be presented. A discussion of the biological
implausibility of backpropagation networks will follow.
Practicals:
Students will work through the explorations in Chapter 5.
- Combined model, task learning and other mechanisms.
The advantages and disadvantages of Hebbian and Error driven learning and
how these different methods of learning may be combined.
Individual
explorations: Students will work through the explorations in Chapter 6.
- The brain and implications for biologically plausible neural
networks.
A broad framework of biologically plausible neural
networks and how this framework relates to brain architecture and function.
- Perception, Vision, Object Recognition and Attention.
From the lower level representations of vision to the higher level of object
recognition. The neural networks considered will be placed within the
context of the human dual route (what-where) visual system.
Individual
explorations: Explorations will be based on the ability of the "what-where"
pathway to influence the network's allocation of attention to spatial
locations (Chapter 8).
Note
This web page provides advance information about a module due
to run in the coming academic year. We believe the details are
accurate at the time of writing but they may be subject to
change.