OS-λ1-ELM: Online sequential λ1-regularized-ELM based on ADMM

• Published in: 2017 6th International Symposium on Advanced Control of Industrial Processes (AdCONIP) pp. 371 - 376
• Main Authors: Dazi Li, Zhiyin Liu, Qibing Jin
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01-05-2017
• Subjects: Algorithm design and analysis; Computational complexity; Convex functions; Learning (artificial intelligence); Neural networks; Nonlinear dynamical systems; Training
• DOI: 10.1109/ADCONIP.2017.7983809

As business data and scientific data become larger and larger, the study of incremental learning algorithms becomes more and more important. Online sequential extreme learning machine (OS-ELM) algorithm is an incremental learning algorithm that can learn data one by one. On the basis of OS-ELM, an online sequential extreme learning machine incremental learning algorithm is proposed based on the λ 1 -regularization (OS-λ 1 -ELM). The proposed method can make use of the original learning results and does not need to re-learn all the data, thus can save time and space resources. By adding λ 1 -regularization, the sparse model can effectively avoid the over-fitting problem. At the same time, alternating direction method of multipliers (ADMM) and the proximity algorithm are used to solve the OS-λ 1 -ELM. The algorithm is deduced into a recursive form, which greatly reduces the computational complexity. Experimental results show that the proposed method has good generalization and robustness.