Forgetting memristor based neuromorphic system for pattern training and recognition

详细信息    查看全文

• 作者：Peijian Zhang^aAuthor Vitae ; Chuandong Li^a ; ^{cdli@swu.edu.cn}Author Vitae ; Tingwen Huang^bAuthor Vitae ; Ling Chen^aAuthor Vitae ; Yiran Chen^cAuthor Vitae
• 关键词：Memristors ; Neuromorphic system ; Crossbar ; Pattern training and recognition
• 刊名：Neurocomputing
• 出版年：2017
• 出版时间：26 January 2017
• 年：2017
• 卷：222
• 期：Complete
• 页码：47-53
• 全文大小：2807 K
• 卷排序：222

文摘

This paper presents a neuromorphic system for mean variance based pattern training and recognition. The system contains a self-learning circuit, a training circuit and a recognition circuit. Memristor model with forgetting effect which has memory ability and forgetting effect simultaneously is applied to simulate forgetting mechanism of neuromorphic system. Different from previous work, which divided training circuit as off line process, here the weight-changing circuit and the recognition part are combined on line for pattern training and recognition. For illustration, the whole neuromorphic system is applied to recognize handwriting number ‘0–9’ on gray images, and simulations verify its effectiveness.