Design and Fabrication of Neural Networks for Signal Processing Recognition

用于信号处理识别的神经网络的设计和制造

• 批准号: 8716099
• 负责人: Shihab Shamma
• 金额: $ 22.78万
• 依托单位: University of Maryland, College Park
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 1988
• 资助国家: 美国
• 起止时间: 1988-08-01 至 1992-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=8716099&HistoricalAwards=false
• 关键词: Design; Fabrication; Neural; Networks; Signal

In prior work, the P.I. has developed and tested a detailed quantitative model of how speech information is processed in the cochlea of mammals. He has studied how this information, after processing, is allocated with speech sounds, as recognized by humans. In this project, the P.I. will go on to apply various learning algorithms to processed human speech, so as to demonstrate the value of this processing in speech recognition applications. Innovative learning algorithms exploiting the true series nature of speed data will be explored, along with more clinical biologically-motivated learning rules.

在之前的工作中，P.I.已经开发并测试了一个详细的量化模型，该模型描述了哺乳动物耳蜗中语音信息是如何处理的。他研究了这些信息在处理后是如何与人类识别的语音一起分配的。在这个项目中，P.I.将继续将各种学习算法应用于处理后的人类语音，以展示这种处理在语音识别应用中的价值。将探索利用SPEED数据的真实序列性质的创新学习算法，以及更多临床生物动机的学习规则。