On the stationary state of topologically ordered competitive learning
Michael D. Lemmon · 2002
The author proposes a competitive learning algorithm for learning nonparametric representations of unknown probability density functions. The proposed algorithm is shown to generate a reversible Markov chain whose invariant distribution is explicitly computed. The computed distribution is used to derive a nonparametric density estimate of unknown density functions. This fact allows the use of the algorithm's representation in estimating the modes of the unknown density function.>