《机械设计基础》试题库及答案.pdf

arXiv:/9707006v1 28 Jul 1997 Learning from mistakes Dante R. Chialvo* ?and Per Bak* ??Niels Bohr Institute, Blegdamsvej 17, Copenhagen, Denmark.?Division of Neural Systems, Memory and Aging, University of Arizona, Tucson, AZ 85724, USA. *Santa Fe Institute, 1399 Hyde Park Rd., Santa Fe, NM 87501, USA. (February 4, 2008) Abstract A simple model of anised learning with no classical(Hebbian) rein- forcement is presented. Synaptic connections involved in mistakes are de- pressed. The model operates at a highly adaptive, probably critical, state reached by extremal dynamics similar to that of recent evolution models. Thus, one might think of the mechanism as synaptic Darwinism. It is widely believed that learning in the brain resides in alterations of synaptic e?cacy. Without exception, contemporary formulations of such learning follows Hebb’s ideas [1] of reinforcement: synaptic connections among neurons excited during a a given ?ring pattern are strengthened by a process of long term potentiation (LTP). However, long term synaptic depression (LTD) in the mammalian brain is almost as prevalent as potentiation, but there appears to be little orno understanding of its functional role. Working hypotheses covers a wide range, where depression is given always an auxiliary function to potentiation [2]. A recent review [3], re?ecting the current variety of ideas regarding the functional role of LTD, speculates: “Although it is conceivable that LTP is the critical phenomena used for storing information, and that LTD may exist simply to reset 1 LTP, it must be noted that it is also conceivable for the converse to be true.” We present an alternative to Hebbian learning. Turning things upside down, we suggest that LTD is, in some instances of learning and development, the fundamental mechanism with LTP playing a secondary role. This view is supported by studies of a simple neuronal learning model. There are two fundamental di?erences between the classical view of learning by rein