基于成员反馈的上下文传播模型

论文

需积分: 5 179 浏览量更新于2024-08-07 收藏 389KB PDF 举报

身份认证购VIP最低享 7 折!

30元优惠券

"Context propagation based on membership feedback" 是一篇由Kohei Inoue和Kiichi Urahama发表在《Systems and Computers in Japan》期刊上的学术论文，主要探讨了基于成员反馈的上下文传播模型在模式识别中的应用。这篇论文的核心内容是提出了一种简单的模式歧视模型，该模型结合了时间或空间上下文，并通过高层胜者全拿（WTA）神经元的成员值反馈到低层模式选择响应神经元。作者们介绍了一种无教师训练方法，基于最大似然原则进行学习。他们的研究旨在模拟人类视觉系统的某些功能，使模式识别过程更加高效和精确。 1. 时间上下文：在时间上下文部分，作者研究了第一时刻的鉴别效果如何反馈到下一时刻的情况。他们展示了聚类是通过呈现时刻的接近性而不是模式相似性来完成的。这意味着系统能根据前后时间点的信息关联性来识别和处理模式，提高了模式识别的连贯性和稳定性。 2. 空间上下文：对于空间上下文，论文指出类似的模式识别方法可以应用于图像模式的空间平滑。这种方法有助于消除噪声和增强图像的局部特征，使得系统能够在空间维度上更好地理解模式，类似于人眼对图像细节的处理。 3. 上下文依赖和成员反馈：上下文依赖是指模式识别过程中的环境信息对其结果的影响。而成员反馈机制允许系统根据高层神经元的决策影响低层神经元的活动，从而实现上下文信息的有效传播。这种机制使得模式识别不完全依赖于输入模式本身，而是利用了周围环境提供的额外信息。 4. 无教师训练：传统的机器学习往往需要有标注的数据进行监督学习。然而，该论文提出的无教师训练方法则不需要预先设定的训练样本，而是基于最大似然原则自动学习，这降低了对大量标注数据的依赖，提高了模型的自适应性。关键词：上下文依赖；成员反馈；拓扑不变模式识别；无教师训练。这篇论文深入研究了如何利用上下文信息和反馈机制来改进模式识别的性能，对于理解人工智能和计算机视觉领域的上下文传播理论有着重要的贡献。同时，提出的无教师训练方法也为实际应用提供了新的可能性。

资源详情

资源推荐

Context Propagation Based on Membership Feedback

Kohei Inoue and Kiichi Urahama

Faculty of Visual Communication Design, Kyushu Institute of Design, Fukuoka, Japan 815-8540

SUMMARY

The authors discuss a simple model for pattern dis-

crimination incorporating temporal or spatial context by

feedback of the membership values output at the high-rank

winner-take-all (WTA) neurons to the lower-rank pattern

selection response neurons. They propose a teacherless

training method based on maximum likelihood. First, for

the temporal context, they study examples with the feed-

back of discrimination effect of a first moment to the next

moment, and demonstrate that clustering is performed by

proximity of presentation moments rather than by the pat-

tern similarity. Next, for the spatial context, they show that

a similar pattern recognition method can be applied to

nica, Syst Comp Jpn, 32(1): 4552, 2001

Key words: Context-dependence; membership

feedback; topologically invariant pattern recognition;

teacherless training.

1. Introduction

Human pattern recognition is affected by the tempo-

ral context and spatial context. Sakai and Miyashita [1]

conducted tests of union tasks on charts to study how they

are affected by the temporal context, and demonstrated that

the charts pattern pairing is formed only by proximity of

presentation moments, rather than by pattern similarity. It

was discovered that such union memory can be represented

by two types of neurons: neurons for encoding and neurons

for recollection. The neurons for recollection are stimulated

only by the context information even if there is no pattern

input. Then, the response of the low-order vision neurons

is also affected by stimuli outside of the reception field,

which demonstrates a spatial context effect [2]. The fill-in

phenomenon shows that response due to surrounding con-

text information is generated even at locations without

input. Based on this physiological and psychological infor-

mation, researchers have tried modeling of learning for

deformation-invariant pattern recognition by using tempo-

ral context [35].

This article relates to simple modeling of the context

effect. The spatial context propagates both by long-distance

lateral connections between neurons and by feedback con-

nections. In this study, we shall discuss exclusively the

feedback effect for both temporal and spatial context. In

other words, it is assumed that in pattern recognition de-

vices the context information is supplied in top-down fash-

ion, to enrich the bottom-up information from the input

patterns. If all of the information is somehow captured in

one mode, it can be considered a variation of the multimo-

dal pattern recognition (with two modes: top-down and

bottom-up). The authors [6] had previously proposed a

teacherless method of learning for close-neighborhood

multimodal pattern recognition. Based on that previously

introduced method, we propose in the current study a

pattern recognition and learning method that involves con-

text. First, we discuss cases relating to the temporal context,

when discrimination effects of the previous moment are fed

back to the next moment. It will be shown that clustering is

based on proximity of presentation rather than on similarity

of patterns, and simple applications to topologically invari-

ant pattern recognition will be demonstrated. Next, it will

be shown for the spatial context that the same pattern

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

Translated from Denshi Joho Tsushin Gakkai Ronbunshi, Vol. J82-D-II, No. 3, March 1999, pp. 494500

下载后可阅读完整内容，剩余7页未读，立即下载

weixin_38622962

粉丝: 3
资源: 903

基于成员反馈的上下文传播模型

基于成员反馈的上下文传播

Jaeger中的Context Propagation机制解析

In neural networks, in the context of error back propagation learning, define learning rate and explain its effect on the learning process.

TX: TXContext.2444090:tag=Implicit-transaction, level=1, propagation=NOT_SUPPORTED, rollback=false, halt=false, ended=false

Transactional(propagation = Propagation.SUPPORTS)

@propagation= Propagation.REQUIRED

@Transactional(propagation = Propagation.REQUIRES_NEW)

propagation

label_propagation模块如何导入

@Transactional中的propagation

@Transactional(propagation = Propagation.REQUIRED, rollbackFor = Exception.class)位置

PROPAGATION_REQUIRES_NEW

propagation=required和supports区别

propagation = propagation.requires_new

方法a上@Transactional(propagation = Propagation.REQUIRES_NEW)，在方法a中有方法b、方法c，如何让a方法报错后，不影响b、c的事务

@transactional中propagation

@transactional(propagation = p

基于tdoa的fang算法 MATLAB

@Transactional(propagation = Propagation.REQUIRED,rollbackFor = {Exception.class})

@Transactional(rollbackFor = Exception.class,propagation = Propagation.REQUIRED

最新资源