Granular computing is a state-of-the-art methodology that simulates the multi-granu-larition cognitive pattern of the human brain to deal with complex problems.As a typical descrip-tion of granular computing,fuzzy quotient space theory focuses on gradually granulating complex problems into the hierarchical multi-granularition spaces,thereby implementing hierarchical solu-tion of the complex problems.However,when dealing with massive high-dimensional data,the efficiency of constructing multi-granularition spaces through the fuzzy similarity relations in the existing fuzzy quotient space methods reduces significantly.On one hand,the fuzzy similarity re-lation is obtained through calculating the similarity among all objects,which is not conducive to processing large datasets;On the other hand,the fuzzy similarity relation contains a large amount of redundant information,which leads to a large number of redundant computation in the subsequent steps.Therefore,based on the 2-nearest neighbor fuzzy relation,an efficient con-struction approach for constructing multi-granularition spaces is proposed,which greatly im-proves the efficiency on the premise of ensuring the performance when facing downstream classi-fication tasks.First,based on thek-nearest neighbor algorithm,ak-nearest neighbor fuzzy rela-tion is proposed,and its key properties are discussed and proven.Second,for the multi-granular-ition spaces construction task,parameter analysis is performed on the k-nearest neighbor fuzzy relation,theoretically proving that when k is taken as k,all effective information in the data space could be included.Then,the number of effective positions in the nearest neighbor and sec-ond nearest neighbor phases are defined.And the algorithm for extracting effective values and ef-fective positions of the fuzzy similarity relation is proposed,the efficiency of constructing multi-granularition spaces is improved by about 75％.Finally,relevant experiments are conducted on 9 UCI datasets,3 UKB datasets,3 image datasets and 3 text datasets to validate the efficiency of multi-granularition spaces construction approach.By comparing and analyzing with the existing classifiers,the effectiveness,stability,and saliency of the proposed approach for classification tasks are demonstrated.In summary,a k-nearest neighbor fuzzy relation that only contains sparse effective information is constructed.On the basis,an efficient construction approach for constructing multi-granularition spaces based on 2-nearest neighbor fuzzy relation is proposed,which greatly reduces time complexity while ensuring classification performance.
关键词
粒计算/多粒度空间/k近邻/模糊关系/模糊商空间
Key words
granular computing/multi-granularition spaces/k-nearest neighbors/fuzzy relation/fuzzy quotient space
引用本文复制引用
基金项目
国家自然科学基金(62276038)
国家自然科学基金(62221005)
重庆市自然科学基金(cstc2019jcyjcxttX0002)
Joint Fund of Chongqing Natural Science Foundation for Innovation and Development(CSTB2023NSCQ-LZX0164)
维普
万方数据