Layered dynamic auto-associative memory with auto-encoder and feedback

Abstract

An aim of this paper is to propose a layered dynamic autwwiociative memory which treats binary patterns as input/output and consists of a (multi-)layered network trained by the backpropagation and a simple feedback from output to input. This simple architecture, auto-encoder type fegdforward nekwork with feedback dynamics, works as an extended version of auto-associative memory, and overcomes poor ability of auto-encoder for recalling a learned pattern. It also solves two critical problems of conventional correlation type associative models: crosstalk noise elimination and multiple-match resolution.In this paper, we show that decision hyperplanes, which are formed in a multilayer feedforward network by means of the backpropagation or LMS ( Least Mean Squared ) learning algorithm, characterize the performance of the proposed model. Abstract: The effects of mapping an ensemble of vectors, using an external coding sequence, upon the retrieval efficiency of such an ensemble in a neural networks model are studied. Computer simulations show an increase in capacity and in the retrieval efficiency by using such a coding. Significant improvements were obtained by using a coded concatenation procedure upon ensembles of highly correlative vectors. Abstract AUTOASSOCIATIVE MEMORY WITH ADJUSTABLE BEHAVIOR Mmfq Rsshid Microelectornic and Computer Tkhnology Corp. Austin, Texar '18759.6509 35M) Weat Balcones Center Dr.N e d networks with high-order interactions only has been shown to be sufficient to provide satisfactory amctivity to the stored patterns and error corrections. Such interactions inmase the storage capacity of the networks and allow to solve a class of problems which were intractable with standad networks. In this paper we analyze the capacity of these higher order networks by the statistical method and show that why the probability of the Abstract We describe an autoassociative high capacity neural network memory architecture based on competitive learning mechanism. This network uses a modified version of ART 1 dynamics and inherits some of its desirable properties. It has a tunable error correction capability and well behaved basins of attraction, which are independent of the size of the network. The dynamic states of neurOnS active and passive can always be chosen with Probability 0.5 each.adjustment of stored vectors according to the idiosyncracies of a particular input environment is another notable feature of this architecture, especially in the absence of a complete knowledge of the nature of the input environment. 11-571

Transfer learning-enabled classification of drugs for Alzheimers Disease

“…Another feature reduction technique employed was the autoencoder. It is a form of unsupervised artificial neural network that reduces the features into fewer dimensions [9]. It had an advantage of selfevaluation, which meant that when the features were reduced they would be expanded again to automatically test if the expanded features fitted the original ones.…”