Learning by random walks in the weight space of the Ising perceptron
Several variants of a stochastic local search process for constructing the synaptic weights of a Isjng perceptron are studied. In this process, binary patterns are sequentinlly inpated to the Ising perceperon and are then learned as the synaptic weight configuration is modified through a chain of single or double-weight flips within the compatible weight configuration space of the earller learned patterns. This process is able to reads a storage capacity of α≈ 0.63 for pattern tength N=101 and α≈ 0.41 for N=1.001. If in addition a relcatning process is exploited, the learning performance is further improved to a storage capacity of α ≈ 0.80 for N=101 and α ≈ 0.42 for N=1001. We found that, for a given learning task, the solutions coustructed by the random walk learning process are separated by a typical Hamming distance, which decreases with the constraint density a of the learning task: at a fixed value of α, the width of the Hamming distance distributions decreases with N.
fluctuation phenomena, random processes, noise, and brownian motion spin-glass and other random models neural networks
Haiping Huang Haijun Zhou
Key Laboratory, of Frontiers in Theoretical Physics. Institute of theoretical Phyeics. Chinese Aca Knvli institate for Theoretical Physies China. Institute of Theorelical Physics, Chinese Academy of
国际会议
International Workshop on Statistical Physics and Computer Sciences(统计物理与计算机科学交叉研究国际研讨会 )
北京
英文
53-60
2010-07-08（万方平台首次上网日期，不代表论文的发表时间）