The Structure Extension Of Associative Memories Based On Data Processing

Recently,with the development of science and technology,the pursuit of event material is getting higher and higher and a large number of data should be recorded with follow shortly.Nowadays,data sets are not only high-dimensional data sets,but also interact with each other,and higher requirements are put forward for the speed,mechanism and efficiency of data processing.More attention has been paid to the research of data processing methods.Studies on the recording,processing and analysis of data have been the essential parts of data processing.This main objective of this dissertation is to introduce a logic-oriented associative memory which constitutes the developments of the architecture,to achieve the storage and the recall mechanisms efficiently,quickly,accurately.This dissertation carries out a series of theoretical researches and developments on associative memories used gradient-based learning mechanisms,particle swarm optimization(PSO)and differential evolution(DE),and the logic development of associative memories is optimized by means of nonlinear transformations(mappings)of spaces of data,autoencoding mechanisms and fuzzy clustering processing.The modification and optimization of these association mechanisms can store,process and recall the data more quickly,efficiently and accurately.These results provide a theoretical basis for the in-depth understanding and development of the big-data processing——granular data,and provide some elemental theoretical basis for further experiments.The main results are as follows:1.A memory matrix is created between the input data set and the output data set.In the single-level fuzzy associative memories,the data are first transformed into a new space and then resulting objects(patterns)are stored in the memory.Then,the recall is obtained by the inverse transformation of the nonlinear transformation.The objective of the transformations is to enhance the recall abilities of the memories.Nonlinear transformations are applied to transform the input space,transform the output space and transform both input and output spaces.The transformations are realized in the form of piecewise linear functions whose parameters are optimized with the use of PSO.A comprehensive suite of experiments involves several types of associative memories such as correlation associative memories,fuzzy associative memories and morphological associative memories.The experiments reveal some interesting relationships between the parameters of the nonlinear mappings and the resulting quality of the recall.They also help quantify the capabilities of the nonlinear mappings to improve the quality of recall.2.Two-level associative memories have one more hidden layer data set than singlelevel associative memories.The first memory matrix is established between the input data and the hidden matrix,and the other memory matrix is established between the hidden matrix and the output data.We develop a logic-driven model of two-level fuzzy associative memories augmented by autoencoding processing.It is composed of two functional modules.The first module of this architecture implements an efficient dimensionality reduction of the original high dimensional data with the use of an autoencoder to get a new representation of the input data.This helps achieve storing and completing the recall realized by a logic-oriented associative memory which constitutes the second module of the architecture.Then,the recall data set is obtained by decoder.The optimization of the association matrices studied in the dissertation involves both gradient-based learning mechanisms and the algorithms of populationbased optimization,i.e.,PSO and DE.A suite of experimental studies is presented to quantify the performance of the proposed approach.With the increase of the number of coding layers and the different nodes in each layer,the data storage and recall capability can be improved to different degrees.3.Associative memories need to have strong capacity to store numerous data sets with a minimal recall error.With the explosion of the database,a huge dataset could be split into some subsets and these subsets interact with each other through a relationship.Tensor associative memories aim at further improvement of the storage capacity and the performance of the recall.In tensor associative memories,we combined with clustering analysis——fuzzy C-means for multiple data sets.The memory matrix comes from membership degree of multiple data sets,which reduces the storage and calculation amount of original associative memory matrix.Then,the recall is restored by using clustering center and membership function.In this part of experiments,particle swarm optimization algorithm and gradient-based method are used to optimize the parameters of the fuzzy C-means.The results show that for tensor associative memory,the combination of fuzzy C-means can effectively improve the data storage capacity and recall with a minimal error.