You need to pick a theme for your main course project. It is expected that you will study some of the related literature and give a presentation in which you dicuss critically one of the relevant papers. You need to implement a simulation to recreate some of the fuindings and possibly investigate some aspects further. Finally, the project findings should be written up in a form of a conference or journal paper. The specific format will be given later. I will assit you individually with your course project, so make sure to talk to me frequently. Just come to my office or email me a time you like to meet.
We need to talk to me individually to decide on the projrct topic. Below are some examples of possible topics and lead papers. To choose a topic you could look at recent issues of the journals Neural Computation, Neural Networks and Neurocomputing and should talk to me directly.
1. Biologically realistic SOMs
Self-organizing maps are a popular clustering method in computer science and were proposed to contribute to the formation of topographic orgasnisations in the cortex. In engeneering applications we use mainly the algorithm proposed by Kohonen, but a biological more direct implementation has been demonstrated a long time ago in a classic paper by Willshaw and Von Der Malsburg. Although there is a wide believe that this mechanisms in what drives cortical organization and reorganization, many basic questions have not been investigated. This could be a rewarding project, but previous students found also that it requires some dedication (time).
Lead paper: D. J. Willshaw and C. Von Der Malsburg, How Patterned Neural Connections Can Be Set Up by Self-Organization, Proceedings of the Royal Society of London. Series B, Biological Sciences, Vol. 194, No. 1117. (Nov. 12, 1976), pp. 431-445.
2. The consequence of STDP on a rate level.
Learning rules in rate models have been well studied theoretically and experimentally. For example, the BCM theory fits well with many experimental observations. However, recently there was new experimental eveidence of spike time dependend palsticity (STDP). The goal of this project is to show how STDP would manifest itself when decribing rates.
Lead paper: Eugene M. Izhikevich and Niraj S. Desai, Relating STDP to BCM, Neural Computation 15, 1511–1523 (2003)
3. Machine learning classification: SVM and ambiguous data
This year I will also allow projects in the area of machine learning as we talked a bit about this in class. There are many good tutorials on SVMs, like
You could explore a specific application of such techniques or study their extension, for example, for classification of ambiguous data
4. Cortico-Hippocampal interactions
5. Interacting recurrent networks
6. Sequence memory
See for example M. Lawrence , T. Trappenberg, A. Fine (2005) Rapid learning and robust recall of long sequences in modular associator networks, Neurocomputing, in press
Other topics include: Storage capacity of continuous attractor networks, sequence memory, etc.