Research On Association Learning Based On Deep Learning

Studying hidden associations between things helps to understand the mechanisms of human cognition and memory.As an important basis of big data analysis,the mining and discovery of data association can effectively help human beings quickly find valuable contents in the face of complex and massive data.At present,most of the existing methods of data association mining are statistical analysis of the existing data,and lack of identification of unknown data.However,there are often many complex nonlinear relations in the data,so it is difficult for the constructive correlation index to describe all the correlation relations.Therefore,from a new perspective,the machine can effectively distinguish the association between objects from the perspective of learning.This paper follows the research frontier at home and abroad,deeply studies and studies the mining methods of association relation,and proposes the mining of association relation from the perspective of learning.Associated learning as one of the important ways of association mining era of big data,different from the traditional correlation intensity measure based on statistical assumption,but using the theory of machine learning,learn by inductive reasoning to known associations with a discriminant function of association discriminant,can the relationship between the unknown object make accurate criterion,the related study results are as follows:(1)With the massive growth of multimodal data in the information age,it is increasingly important to find potential hidden relationships in these data.In order to effectively mine complex association relations in data,this paper proposes the basic framework of relevance learning theory based on the content of machine learning computational learning theory,and defines the input space,feature space,output space,joint probability distribution and hypothesis space of relevance learning.The formalization gives the mathematical expression of relevance learning and defines the criterion of relevance learning,which provides the theoretical basis for the design and implementation of the model of relevance learning and the basic criterion for selecting the optimal model.(2)According to the content of association learning theory,this paper constructs two class associated image data sets(TAID)and constructs the association learning model by taking the classification relationship between animals and fruits as an example.Specifically,the model makes use of the advantages of image feature extraction by using convolutional neural network to extract association features from TAID data,and uses the extracted association features to distinguish association relations from two perspectives,softmax function and k-nearest neighbor algorithm.In this paper,an associated image convolutional neural network discriminator(AICNN)designed by convolution network,an associated image Le Net discriminator(AILe Net)obtained by modifying Le Net network,and an associated image K near-neighbor discriminator(AIKNN)considered from the perspective of traditional machine learning method k-nearest neighbor algorithm are presented.The association discriminators in the three association learning models were trained and tested on the TAID data set respectively,and the test accuracy reached about 85%,which fully demonstrated the rationality of the method of association mining from the perspective of learning and the feasibility of association learning.In short,from the perspective of learning,this paper innovatively puts forward the method of mining association relations,defines the relevant concepts and formal expressions of association learning,trains the association learning model based on the feature extraction method of deep neural network,and obtains an effective association discriminator to distinguish unknown association relations.This paper scientifically explains the feasibility of the relevance learning method,provides a new perspective for other researchers in the field of relevance mining,and provides a feasible method for exploring the potential relevance between unknown things.