Multi-attention concept-cognitive learning model: A perspective from conceptual clustering

Concept-cognitive learning (CCL), as a cognitive process, is an emerging field of simulating the human brain to learn concepts in the formal context. Simultaneously, attention is a core property of all perceptual and cognitive operations. Nevertheless, no current existing CCL models and conceptual clustering methods consider the impact of attention. In light of these observations, in this article, we present a novel concept learning method, called the multi-attention concept-cognitive learning model (MA-CLM), to address the issue by exploiting graph attention and the graph structure of the concept space. This model is deployed toward the goal of conceptual cognitive more reasonable: generate pseudo-concept with higher expected utility while taking into consideration making classification tasks more efficient. Specifically, a conceptual attention space is learned for each decision class via attribute attention. Furthermore, a new concept clustering and concept generation method based on graph attention was proposed based on the conceptual attention space. Comparative studies with S2CL(alpha) over a total of nine UCI data sets validate the effectiveness and efficiency of concept clustering based on graph attention in concept-cognitive learning. In addition, we also performed a comparative evaluation of MA-CLM against several classical classification algorithms to demonstrate the excellent properties in classification tasks. Finally, the model is validated by concept generation on the handwritten numeral dataset MNIST. (c) 2022 Elsevier B.V. All rights reserved.

Automated summary

This article introduces a novel concept learning method, called the multi-attention concept-cognitive learning model (MA-CLM). The model addresses the issue of the impact of attention in current concept-cognitive learning models by utilizing graph attention and the graph structure of the concept space. Experimental results validate the effectiveness and efficiency of concept clustering based on graph attention in concept-cognitive learning, and comparative evaluation with classical classification algorithms demonstrates the excellent properties of the model in classification tasks.