Studies on Multimodal Information Processing Based Upon Fast Expectation-Maximization

基于快速期望最大化的多模态信息处理研究

• 批准号: 11680401
• 负责人: MATSUYAMA Yasuo
• 金额: $ 2.3万
• 依托单位: Waseda University
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for Scientific Research (C)
• 财政年份: 1999
• 资助国家: 日本
• 起止时间: 1999 至 2000
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/grant/KAKENHI-PROJECT-11680401/
• 关键词: multimodal information processing; fast algorithm; α-EM algorithm; α-ICA algorithm; fMRI for brains; α-ICAアルゴリズム; 期待値最大化; 高速ICA; 感性処理処理

This project had the following two targets:(1) Investigation of new algorithms which give optimal structures measured by probabilistic and statistical performance,(2) Applications of the obtained algorithms to multimodal information sources which correspond to human signals.The first year was used to create a new class of information processing methods. In this phase, the following results were obtained:(a) A new class of expectation-maximization algorithm was found. This method was named the α-EM algorithm. The α-EM algorithm contains the traditional log-EM algorithm as a special case. The performance in speed outperforms the traditional log-EM method. This work received the Telecommunications Advancement Foundation Award.(b) The above method using the extended logarithm was found to be applicable to the independent component analysis which separates unknown source signals. This new method was named the α-ICA.In the last year, the above methods (a) and (b) were applied to multimodal information processing. Obtained results are(i) Motion estimation from optical flows,(ii) Estimation of living human brains' activities from functional magnetic images. It was found that there is an active area in the rear of the right hemisphere. This active area is asymmetric.As is explained above, this research project was ended with lots of viable results.

该项目具有以下两个目标：（1）对新算法进行的研究，这些算法提供了通过概率和统计性能衡量的最佳结构，（2）获得的算法对与人类信号相对应的多模式信息源的应用。第一年用于创建新的信息处理方法。在此阶段，获得了以下结果：（a）发现了一种新的期望最大化算法。此方法α-EM算法包含传统的日志EM算法作为特殊情况。速度的性能优于传统的日志EM方法。这项工作获得了电信促进基金会奖。（b）上述使用扩展对数的方法适用于将未知源信号分开的独立组件分析。这种新方法被命名为α-ica。去年，上述方法（a）和（b）应用于多模式信息处理。获得的结果是（i）来自光流的运动估计，（ii）从功能性磁性图像中对活人大脑活动的估计。发现右半球的后部有一个活动区域。该活性区域是不对称的。如上所述，该研究项目以许多可行的结果结束。

项目成果

Y.Matsuyama,et al.: "Fast Learning by the α-ECME Algorithm"Proc.ICONIP'99. 3. 1184-1190 (1999)

Y.Matsuyama 等人：“通过 α-ECME 算法进行快速学习”Proc.ICONIP99。3. 1184-1190 (1999)

The α-EM algorithm and its basic properties

α-EM算法及其基本性质

• 发表时间: 1999
• 期刊: Trans.IEICE J-82-D-I
• 作者: Y.Matsuyama;et al.;Y.Matsuyama
• 通讯作者: Y.Matsuyama

Multiple descent cost competitive learning and data-compressed 3-D morphing

• DOI: 10.1109/iconip.1999.844017
• 发表时间: 1999-11
• 期刊: ICONIP'99. ANZIIS'99 & ANNES'99 & ACNN'99. 6th International Conference on Neural Information Processing. Proceedings (Cat. No.99EX378)
• 作者: Y. Matsuyama;T. Shimazu;G. Matsuo;T. Arisaka

The α-ICA algorithm

α-ICA 算法

• 发表时间: 2000
• 期刊: Proc.2nd Int.Workshop on ICA and BSS
• 作者: Y.Matsuyama;et al.
• 通讯作者: et al.

Y.Matsuyama: "The α-EM algorithm and its applications"Proc.ICASSP-2000. 1. 592-595 (2000)

Y.Matsuyama：“α-EM 算法及其应用”Proc.ICASP-2000。1. 592-595 (2000)。