Concept-cognitive Learning Model Based On Attention

Attention plays an important role in perceptual and cognitive operations.Conceptualcognitive learning(CCL)is an emerging field of research that simulates the conceptual mechanism of human brain learning to construct specific cognitive theories and models,in which attention will play an important role.However,existing CCL models and conceptual clustering methods do not take into account the effects of attention.In order to introduce attention into the CCL model,thesis proposes a series of CCL models based on attention.(1)A Multi-attention concept-cognitive learning model(MA-CLM)based on classical formal context is proposed.With the introduction of the attention mechanism,the concept storage space and concept learning mechanism will change.Specifically,the conceptual attention space is constructed by using the similarity between conditional attributes and decision attributes as a measure of the attention degree of each conceptual space to conditional attributes.Based on the conceptual attention space,a method of concept clustering and concept generation based on multi-attention is proposed.The effectiveness of the multi-attention conceptual clustering method in conceptual cognitive learning is verified by comparing it with the static model of semi-supervised conceptual learning model S2 CL on 9 UCI datasets.In addition,we also compare MA-CLM model with several classical classification algorithms and prove MA-CLM’s excellent performance in classification task.Finally,the concept generation effect of the model is verified on MNIST.(2)The Graph attention fuzzy concept-cognitive learning model(GA-FCLM)is proposed.The MA-CLM model based on the classical formal background is mainly for categorical data,but there is a lot of continuous data in real life.To process continuous data with MA-CLM,the continuous data needs to be converted into a binary formal context through a data preprocessing algorithm,which means that the conversion process will be increased,and the discretization method will have an impact on the model effect.Based on the above analysis,thesis proposes a cognitive learning method of fuzzy concepts based on graph attention.Firstly,some definitions and properties related to GA-FCLM are given based on MA-CLM model.Secondly,the corresponding learning algorithm and concept classification criterion are designed.Finally,the performance of GA-FCLM is verified by comparison experiment of classification accuracy and number of concepts.(3)The Graph attention fuzzy concept-cognitive learning dynamic model(GA-FCLDM)is proposed.GA-FCLM is a static model that cannot accommodate the dynamic changes of the dataset over time,so the conceptual space needs to be reconstructed every time new data is available.Therefore,thesis designs a conceptual dynamic update mechanism based on GA-FCLM model,which is suitable for dynamic data.The experimental results show that the dynamic learning effect of this model is remarkable on some data.