Associative Memory with Pattern Analysis and Synthesis by a Bottleneck Neural Network(Contribution to 21 Century Intelligent Technologies and Bioinformatics)

Eiichi INOHIRA, Takeshi OGAWA, Hirokazu YOKOI
2008 International journal of biomedical soft computing and human sciences  
Abstraet: We propose a new associative memory to improve its noise tolerance and storage capacity. Our underlying model is an improved multidirectional associative memory (IMAM), which uses autoassociative bottlene ¢ k neural networks to remove noise in its input, i.e., analyze pattems. IMAM has inethcient storage capacity and low noise tolerance due to a correlation matrix representing association. One of our basic ideas is te replace a correlation matrix with a multilayer perceptron (MLP),
more » ... erceptron (MLP), which has better learning and generalizatien capability Moreover, we introduce two improvements. One is to add intermediate elements into MLP to improve its perfbmmce. The other is to use outputs of hidden layers in a five-layer bottleneck neural network. These outputs include informatien on synthesis of a key pattern from compressed information in the middle layer. Tb evaluate the proposed approaches, we compared three types of associative memory: associative memory with a bottleneck neural network and MLP (AMfB-M), AMIB-M with intermediate elements (AMfB-I), and Ama-I with synthetic outputs (AM/B-IS ). 10-by-10 images ofLatin alphabet are used as patterns for association, In a case of association between 78 non-iajective pattern pairs with 10% noise, our proposed AMfB-IS is better thari AMIB-M by more than 40% in pattern recalling ratio, Kaywords: Associative memory, bottleneck intermediate element, noise toleranceneural network, multilayer perceptron,
doi:10.24466/ijbschs.13.2_27 fatcat:zgsfx2xjlzdfxpekffxsf7c5pa