提升异联想记忆容量：自适应快速学习算法

需积分: 15 142 浏览量更新于2024-08-07 收藏 184KB PDF 举报

本文档探讨了"联想记忆的自适应快速学习"（Adaptive Quick Learning for Associative Memory）这一主题，由Tomoshige Yoshihara和Masafumi Hagiwara两位学者在Keio大学，日本横滨，223-8522发表。关联记忆（Bidirectional Associative Memory, BAM）是一种异联结型记忆模型，它具备召回和恢复模式的能力。然而，作为基于赫布学习（Hebbian learning）的记忆系统，其主要缺点是存储容量较低。论文提出了一个名为Pseudo-Relaxation Learning Algorithm (PRLAB) 的训练方法，这种方法显著提高了BAM的记忆容量。通过将赫布学习与快速学习算法相结合，BAM的存储能力得到增强，同时对噪声输入的抗干扰性也有所提升，但前提是学习过程中连接权重的更新能够收敛。如果在学习过程中遇到了无法找到解决方案的情况，即训练模式之间的不等式冲突可能导致学习无法收敛，传统方法下的模式回忆无法得到保障。为了解决这个问题，论文提出了一种创新的方法。当系统估计到学习过程可能不会收敛时，通过将随机数附加到训练模式上，实现了多模态化。这样即使在训练模式中存在同时性的不等式矛盾，也能通过这种策略促使学习过程趋于稳定，确保了训练模式的回忆功能。这种方法不仅解决了原有问题，还提高了系统的整体性能和稳定性。通过这一改进，作者们旨在推动关联记忆的学习算法朝着更高效、可靠的方向发展，为未来的神经网络设计和人工智能应用提供了新的思路和技术支持。

Adaptive Quick Learning for Associative Memory

Tomoshige Yoshihara and Masafumi Hagiwara

Keio University, Yokohama, Japan 223-8522

SUMMARY

Bidirectional associative memory (BAM) is a a form

of heteroassociative memory that can recall and restore

patterns. Being a Hebbian learning-based memory, it has

the problem of very low capacity. A training paradigm

called the Pseudo-Relaxation Learning Algorithm

(PRLAB) greatly increases the memory capacity of BAM.

The combination of Hebbian learning with BAM in the

Quick Learning training algorithm has increased its storage

capacity and robustness to noisy inputs while greatly reduc-

ing the number of iterations. But in these learning algo-

rithms, if a solution domain does not exist for the set,

learning of the connection weights will not converge and

recall of the training pattern is not guaranteed. This paper

proposes a new method of solving this problem, in which

training patterns are multimodalized by attaching random

numbers to them if it is estimated that learning is not

converging. Thus even if there is a contradiction in the

simultaneous inequalities used in the training patterns, con-

vergence is artificially forced and correct recall becomes

possible. Simulations indicate the effectiveness of the new

method in both the presence and absence of untrainable

5361, 2001

Key words:

Bidirectional associative memory

(BAM); pseudo-relaxation learning algorithm (PRLAB);

Hebbian learning; unrainable patterns; multimodalization;

forced convergence.

1. Introduction

Memory is the most important function of the brain,

and therefore understanding its mechanism is very im-

portant. In order to study it, many researches based not

only on physiological and psychological approaches, but

also engineering approaches have been made [15].

Among them, researches based on neural models have

been reported [113]. Bidirectional Associative Memory

(BAM) is a simple structure that is built taking notice of

similarities with the brain. BAM consists of two layers

with a symmetric synaptic interconnection between lay-

ers. It behaves as a heteroassociative memory that can

recall and store patterns.

However, as most associative recalling memories,

BAM is also a Hebbian learning-based memory and has the

problem of very low memory capacity. To improve the

storage capacity of BAM, several improvements have been

proposed [213]. The multiple training technique [7] can

improve the storage capacity of BAM but does not guaran-

tee recall of all training patterns. Wang and colleagues

method [8], on the other hand, guarantees the recall of

all training patterns, but the storage capacity is still low.

Kawabata and colleagues [13] proposed an algorithm to

improve the storage capacity and processing time

through multimodalization of the training vectors. How-

ever, since correlation-based learning is used, the storage

capacity is still low.

A new training paradigm called Pseudo-Relaxation

Learning Algorithm for BAM (PRLAB) was proposed by

Systems and Computers in Japan, Vol. 32, No. 1, 2001

Translated from Denshi Joho Tsushin Gakkai Ronbunshi, Vol. J82-D-II, No. 1, January 1999, pp. 109116

Contract grant sponsor: Japan Society for the Promotion of Science;

contract grant #JSPS-RFTF 96 I 00102.

下载后可阅读完整内容，剩余8页未读，立即下载

weixin_38720997

粉丝: 7
资源: 888

提升异联想记忆容量：自适应快速学习算法

神经网络 联想记忆

基于Hopfield网络的PID模型参考自适应控制.rar_hopfield网络_pid参考模型_网络控制_自适应PID_自适应

BP神经网络模型与学习算法

(完整版)第七章机器学习.docx

LM优化的半自适应遗传神经网络在入侵检测中的应用

自组织神经网络与竞争学习算法详解

自组织神经网络与竞争学习算法原理

清华大学神经网络课程：学习率α与训练策略

Spring Boot Starter-kit：含多种技术应用，如数据库、认证机制，有应用结构.zip

包含 Spring Boot 等系列技术参考指南中文版及相关资源的仓库.zip

最新资源

神经网络联想记忆