Superimposing Learning for Backpropagation Neural Networks | 重ね書き学習ニューラルネット

Bull. Coll. Med. Sci. Tohoku Univ. 11(3): 253～259，2002

Superimposing Learning for Backpropagation
Neural Networks

Noriyasu HOMMA^*,** and Madan M. GUPTA^**
*Department of Radiological Technology, College of Medical Sciences, Tohoku University
^**Intelligent Systems Research Laboratory, College of Engineering, University of Saskatchewan

重ね書き学習ニューラルネット

本間経康^*,**，Madan M. GUPTA^**
^*東北大学医療技術短期大学部診療放射線技術学科
^**サスカッチワン大学工学部知的システム研究所

Key words : Neural Networks, Learning, Structural adaptation, Long-term memory,
Short-term memory, Incremental learning, Backpropagation

In this paper a new neural network model is presented for incremental learning tasks where networks are required to learn new knowledge without forgetting the old one. An essential core of the proposed network structure is their dynamic and spatial changing weights (DSCWs). A learning scheme is developed for the formulation of the dynamic changing weights, while a structural adaptation is formulated by the spatial changing connecting weights. To avoid disturbing the past knowledge by the creation of new connections, a restoration mechanism is introduced by using the DSCWs. Usefulness of the proposed model is demonstrated by using a system identification task.