扑墓技载指炯.PDF

NSL-000905. Septemb er, 2000 To b e published in Neural Networks, 14 (2001) Mo dularity, Evolution, and the Binding Problem: A View from Stability Theory Jean-Jacques E. Slotine and Winfried Lohmiller Nonlinear Systems Lab oratory Massachusetts Institute of Technology Cambridge, Massachusetts, 02139, USA jjs@mit.edu, wslohmil@mit.edu Abstract Any biological ob ject, and sp ecically the brain, is the result of evolution. Evo- lution pro ceeds by accumulation and combination of stable intermediate states as is well known, survival of the ttest really means survival of the stable. Simple ex- amples ab ound: for instance, human emotional resp onse involves b oth a fast archaic lo op bypassing the cortex, and a slower cortical lo op; motion control architecture in vertebrates is b elieved to involve combinations of simple motor primitives. However, in themselves, accumulations and combinations of stable elements have no reason to b e stable. Hence the hyp othesis that evolution will favor a particular form of stabil- ity, which automatically guarantees stability in combination. Such a form of stability, which we refer to as "contraction," can b e characterized mathematically. Thus, con- traction theory may help guide functional mo delling of the central nervous system, and conversely it provides a systematic metho d to build arbitrarily complex rob ots out of simpler elements. Furthermore, contraction theory may shed light on the problem of p erceptual un