Analysing Cluster Formation by Replica Method

Using the replica method, we derive quantitative information about the clustering into 2 subgroups of data lying on the surface of an N-dimensional hypersphere. Various clustering algorithms are considered, including the one corresponding to competitive learning.

[1]  M. Opper,et al.  Maximal stability in unsupervised learning , 1995 .

[2]  Van den Broeck C,et al.  Gradient descent learning in perceptrons: A review of its possibilities. , 1995, Physical review. E, Statistical physics, plasmas, fluids, and related interdisciplinary topics.

[3]  Andreas Engel,et al.  Uniform Convergence Bounds for Learning from Examples , 1994 .