The competitive learning vector quantization (CLVQ) algorithm with
constant step $\varepsilon > 0$--also known as the Kohonen algorithm with 0
neighbors--is studied when the stimuli are i.i.d. vectors. Its first
noticeable feature is that, unlike the one-dimensional case which has $n!$
absorbing subsets, the CLVQ algorithm is "irreducible on open sets"
whenever the stimuli distribution has a path-connected support with a nonempty
interior. Then the Doeblin recurrence (or uniform ergodicity) of the algorithm
is established under some convexity assumption on the support. Several
properties of the invariant probability measure $\nu^{\varepsilon}$ are
studied, including support location and absolute continuity with respect to the
Lebesgue measure. Finally, the weak limit set of $\nu^{\varepsilon}$ as
$\varepsilon \to 0$ is investigated.
References
[1] Benveniste, A., M´etivier, M. and Priouret, P. (1987). Algorithmes Adaptatifs et Approximations Stochastiques. Masson, Paris.
[2] Bouton, C. and Pag es, G. (1993). Self-organization and convergence of the one-dimensional Kohonen algorithm with nonuniformly distributed stimuli. Stochastic Process. Appl. 47 249-274.
[3] Bouton, C. and Pag es, G. (1994). Convergence in distribution of the 1-dimensional Kohonen algorithms when the stimuli are not uniform. Adv. in Appl. Probab. 26 80-103.
[4] Bouton, C. and Pag es, G. (1994). Convergence in distribution of the multi-dimensional Kohonen algorithm with 0 neighbor. Technical Report 259, Univ. Paris 6.
[5] Burton, R. M. and Faris, W. G. (1996). A self-organizing cluster process. Ann. Appl. Probab. 6 1232-1247.
[6] Cohort, P. (1997). Unicit´e d'un quantifieur optimal par le th´eor eme du col. Technical report, Univ. Paris 6.
[7] Cottrell, M. and Fort, J. C. (1987). Etude d'un algorithme d'auto-organisation. Ann. Inst. H. Poincar´e S´er. B 23 1-20.
[8] Doob, J. L. (1953). Stochastic Processes. Wiley, New York.
[9] Duflo, M. (1990). M´ethodes R´ecursives Al´eatoires. Masson, Paris.
[10] Flanagan, J. A. (1994). Self-organizing neural networks. Ph.D. dissertation, Ecole Poly technique F´ed´erale de Lausanne.
[11] Fort, J. C. and Pag es, G. (1995). On the a.s. convergence of the Kohonen algorithm with a general neighborhood function. Ann. Appl. Probab. 5 1177-1216.
[12] Fort, J. C. and Pag es, G. (1996). Asy mptotics of the invariant distribution of a stochastic algorithm with constant step. Technical Report 324, Univ. Paris 6.
[13] Gray, R. M. ed. (1982). Special issue on quantization. IEEE Trans. Inform. Theory 28 127- 394.
[14] Kieffer, J. C. (1982). Exponential rate of convergence for Lloy d's method I. IEEE Trans. Inform. Theory 28 205-210.
[15] Kohonen, T. (1982). Analy sis of a simple self-organizing process. Biol. Cy bernet. 44 135- 140.
[16] Kohonen, T. (1984). Self-Organization and Associative Memory. Springer, New York.
[17] Kohonen, T. (1991). Self-organizing maps: optimization approaches. In Artificial Neural Networks (T. Kohonen, et al., eds.) 981-990. North-Holland, Amsterdam.
[18] Kushner, H. J. and Clark, D. S. (1978). Stochastic Approximation for Constrained and Unconstrained Sy stems. Springer, New York.
[19] Lamberton, D. and Pag es, G. (1996). On the critical points of the one-dimensional competitive learning vector quantization algorithm. Proceedings ESANN 1996 97-102.
[20] Mey n, S. P. and Tweedie, R. L. (1993). Markov Chains and Stochastic Stability. Springer, New York.
[21] Pag es, G. (1993). Voronoi tessellation, space quantization algorithms and numerical integration. Proceedings ESANN 1993 221-228.
[22] Revuz, D. (1975). Markov Chains. North-Holland, Amsterdam.
[23] Sarzeaud, O., St´ephan, Y. and Touzet, C. (1990). Application des cartes auto-organisatrices a la g´en´eration de maillages aux ´el´ements finis. Proceedings of Neuro-N imes 1 81-96.
[24] Trushkin, A. V. (1982). Sufficient conditions for uniqueness of a locally optimal quatizer for a class of convex error weighting functions. IEEE Trans. Inform. Theory 28 187-198.