| Subject area |
Data analysis |
| Question |
Is there room for improvement on
existing clustering techniques for quantative
data? |
| Hypothesis |
n/a - literature study |
| Conclusion |
It looks as though there may be a range
of possible
improvements. |
References
|
Method :
Using existing literature, investigate the preferred
methods of clustering data. Using thought experiments,
produce potential ways of improving existing
algorithms.
Ethics :
No ethical considerations.
Risk analysis:
No risks associated.
Time scale:
Two weeks literature review and fast prototyping for
candidate solutions |
Results:
Potential benefits can be seen in using Kohonen networks
with :
- zero initialisation
- wrapping target map
- kernel functions on inputs
- weighted updates
Possible reuse of other existing technologies :
- Kernel engines
- Climate Space Modelling
- alternative geometry with mean tracking (Mitchell, RJ)
- modify Stochastic Diffusion Search (Bishop,
M)
|
