Active Semi-supervised Transfer Learning (ASTL) for Offline BCI Calibration

Dongrui Wu

DOI: https://doi.org/10.48550/arXiv.1805.05781

2018-05-12

Abstract:Single-trial classification of event-related potentials in electroencephalogram (EEG) signals is a very important paradigm of brain-computer interface (BCI). Because of individual differences, usually some subject-specific calibration data are required to tailor the classifier for each subject. Transfer learning has been extensively used to reduce such calibration data requirement, by making use of auxiliary data from similar/relevant subjects/tasks. However, all previous research assumes that all auxiliary data have been labeled. This paper considers a more general scenario, in which part of the auxiliary data could be unlabeled. We propose active semi-supervised transfer learning (ASTL) for offline BCI calibration, which integrates active learning, semi-supervised learning, and transfer learning. Using a visual evoked potential oddball task and three different EEG headsets, we demonstrate that ASTL can achieve consistently good performance across subjects and headsets, and it outperforms some state-of-the-art approaches in the literature.

Machine Learning,Human-Computer Interaction

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to reduce the amount of individual - specific calibration data required for single - trial event - related potential (ERP) classification in brain - computer interface (BCI) systems. Due to large individual differences, it is usually necessary to collect some specific labeled data for each subject to adjust the classifier. However, this individual - specific data collection process is both time - consuming and labor - intensive. Therefore, the paper proposes an active semi - supervised transfer learning (ASTL) method for offline BCI calibration, aiming to use some unlabeled auxiliary data to reduce the amount of required labeled data, thereby improving the performance of the classifier.

Specifically, the paper considers a more general situation where some of the auxiliary data may be unlabeled. By integrating active learning (AL), semi - supervised learning (SSL) and transfer learning (TL), ASTL can achieve consistently good performance across different subjects and different EEG headsets, and outperforms some of the state - of - the - art methods in the literature.