X-means: Extending K-means with Efficient Estimation of the Number of Clusters
ICML '00 Proceedings of the Seventeenth International Conference on Machine Learning
| Hi-index | 0.00 |
The dissociated rat hippocampal neurons on a multielectrodes array dish are useful as simple model of brain information processing system. We analyzed spontaneous activity in the living neuronal network to investigate periodicity and stability of neuronal network activity. Electrical activity pattern at 5 ms time window was represented as a feature vector with 64 elements 0 or 1, corresponding to presence or absence of spike detected at each electrode. X-means clustering method with kkz algorithm preprocessing was applied to the feature vector of each time window. The number of clusters was stable for 30 min with some fluctuations. As extending of clustering range from 5 min to 30 min in 5 min increments, the estimated number of cluster increased, suggesting the number of activity patterns was not stable and increase. However, highly reproducible clusters were stable against extension of clustering range. In addition, the number of highly reproducible clusters was saturated at approximately for 40 s clustering range. These results suggested that the spike patterns compose limited number of highly reproducible clusters and a lot of small clusters derived from reproducible clusters, and highly reproducible clusters were expressed repeatedly. Semi-artificial neuronal network possessed pattern repertories and they are considered to be able to express certain states.