文献类型：期刊文献

英文题名：Small-sample size problems solving based on incremental learning: an adaptive Bayesian quadrature approach

作者：Feng, Yiding[1,2];Feng, Xiang[1,2];Yu, Huiqun[1,2]

机构：[1]East China Univ Sci & Technol, Dept Comp Sci & Engn, Shanghai 200237, Peoples R China;[2]Shanghai Engn Res Ctr Smart Energy, Shanghai 200237, Peoples R China

年份：2023

卷号：53

期号：12

起止页码：15174

外文期刊名：APPLIED INTELLIGENCE

收录：;EI(收录号：20224613114201);WOS:【SCI-EXPANDED(收录号:WOS:000881894700003)】；

基金：This work was supported in part by the Key Program of National Natural Science Foundation of China under Grant No.62136003, the National Natural Science Foundation of China under Grant No.62276097, Shanghai Economic and Information Commission "Special Fund for Information Development" under Grant No.XX-XXFZ-02-20-2463, Scientific Research Program of Shanghai Science and Technology Commission under Grant No.21002411000.

语种：英文

外文关键词：Incremental learning; Bayesian quadrature; Small-sample size problems; Multi-startADAM; Programming problems

摘要：When solving programming problems with objectives, we are often faced with the challenge of insufficient samples. And when new samples are generated, re-modeling based on historical and incremental samples is costly. Many Bayesian based approaches have been proposed to update the original models using newly coming samples which are called incremental learning. In this paper, we derive a calculation method of information value based on Bayesian quadrature technology, and use multi-start Adaptive Momentum (ADAM) stochastic gradient algorithm to design independent adaptive learning rates by computing the first-order moment and the second-order moment of the gradient then we propose a small-sample size problems solving based on incremental learning called Adaptive Bayesian Quadrature Approach (ABQA) We use a numerical validation experiment to verify the effectiveness of our approach, and then use two programming problems, shared bike system and newsvendor problem, to verify the effectiveness of our approach on incremental learning and small-samples size problems. In all these experiments, ABQA outperforms the benchmarks and state-of-the-art Bayesian algorithms. Performance and speed are improved by 2.8% and 8.6%, respectively.