Title :
An Improved Algorithm for Incremental Learning LEARN++
Author_Institution :
Shanxi Univ. of Finance & Econ., Taiyuan
Abstract :
An Improved Algorithm for Incremental Learning LEARN++ is proposed. Based on the LEARN++ algorithm, The Improved Algorithm modifies instances selection technique in Boosting and provides an effective method of selecting the individual networks. The Algorithm allows the new classifiers trained with new database to learn past knowledge reducing the forgetfulness , and ensures generalization performance by selecting individual networks. The experiments show that the performance of Improved Algorithm is better than that of LEARN++ and its memory capacity and computing cost. are acceptable.
Keywords :
generalisation (artificial intelligence); learning (artificial intelligence); LEARN++; generalization performance; improved algorithm; incremental learning; memory capacity; Algorithm design and analysis; Boosting; Costs; Databases; Finance; Pattern analysis; Pattern recognition; Testing; Voting; Wavelet analysis; Boosting; Incremental learning; LEARN++;
Conference_Titel :
Wavelet Analysis and Pattern Recognition, 2008. ICWAPR '08. International Conference on
Conference_Location :
Hong Kong
Print_ISBN :
978-1-4244-2238-8
Electronic_ISBN :
978-1-4244-2239-5
DOI :
10.1109/ICWAPR.2008.4635795