A chunk updating LS-SVMs based on block Gaussian elimination method

作者：Song, Xiaoxin; Jian, Ling; Song, Yunquan^*

来源：Applied Soft Computing, 2017, 51: 96-104.

DOI：10.1016/j.asoc.2016.12.004

• Incremental online learning
• Chunk updating
• LS-SVMs
• Blast furnace ironmaking process
• Tendency predictiona

Existing online learning LS-SVMs algorithms incorporate support vectors (SVs) one-by-one. Therefore, it limits the usage of these algorithms in applications where data arrives in a fashion of mini-batch, i.e., several samples arrive at the same time. In this paper, an IOLS-SVMs algorithm which incorporates SVs-chunk-by- chunk is proposed for online learning. Each time when a new chunk of SVs are incorporated into the SVs set, the proposed IOLS-SVMs algorithm employs the block Gaussian elimination method to dynamically update LS-SVMs model. Compared with the state-of-the-art online learning LS-SVMs algorithm, the proposed IOLS-SVMs one can reduce O(nm(2)) operations per iteration with n original and m new incorporated SVs. Experimental results on benchmark and real-world datasets show the validity and efficiency of the proposed algorithm.

• 出版日期2017-2
• 单位中国石油大学（北京）