Domain Adaptation Transfer Extreme Learning Machines

Abstract

Extreme learning machines (ELMs) have been confirmed to be efficient and effective learning techniques for pattern recognition and regression. However, ELMs primarily focus on the supervised, semi-supervised and unsupervised learning problems in single domain and the generalization ability in multiple domains based learning issues is hardly studied. This paper aims to propose a unified framework of ELMs with domain adaptation and improve their transfer learning capability in cross domains without loss of the computational efficiency of traditional ELMs. We integrate domain adaptation into ELMs and two algorithms including source domain adaptation transfer ELM (TELM-SDA) and target domain adaptation transfer ELM (TELM-TDA) are proposed. For insight of the difference among ELM, TELM-SDA and TELM-TDA, two remarks are provided. Experiments on the popular sensor drift big data with multiple batches in machine olfaction, the results clearly demonstrate the characteristics of the proposed domain adaptation transfer ELMs that they can not only copy with sensor drift efficiently without cumbersome measures comparable to state-of-the-art methods but also bring new perspectives for ELM.