Spectral Transfer Learning Using Information Geometry for a User-Independent Brain-Computer Interface.

Nicholas R Waytowich,Addison W Bohannon,Vernon J Lawhern,Kenneth R Ball,Brent J Lance

doi:10.3389/fnins.2016.00430

Abstract

Recent advances in signal processing and machine learning techniques have enabled the application of Brain-Computer Interface (BCI) technologies to fields such as medicine, industry, and recreation; however, BCIs still suffer from the requirement of frequent calibration sessions due to the intra- and inter-individual variability of brain-signals, which makes calibration suppression through transfer learning an area of increasing interest for the development of practical BCI systems. In this paper, we present an unsupervised transfer method (spectral transfer using information geometry, STIG), which ranks and combines unlabeled predictions from an ensemble of information geometry classifiers built on data from individual training subjects. The STIG method is validated in both off-line and real-time feedback analysis during a rapid serial visual presentation task (RSVP). For detection of single-trial, event-related potentials (ERPs), the proposed method can significantly outperform existing calibration-free techniques as well as outperform traditional within-subject calibration techniques when limited data is available. This method demonstrates that unsupervised transfer learning for single-trial detection in ERP-based BCIs can be achieved without the requirement of costly training data, representing a step-forward in the overall goal of achieving a practical user-independent BCI system.

Highlights

Brain-Computer Interfaces (BCIs) are augmentative devices that decode user intent directly from the brain (Wolpaw et al, 2002)
This paper aims to eliminate the need for calibration and develop a user-independent BCI by proposing a new method, Spectral Transfer with Information Geometry (STIG), which leverages an ensemble of information geometric classifiers coupled with spectral-meta learning (SML), an unsupervised ensemble method for inferring the appropriate weights for a linear combination of classifiers (Barachant et al, 2012; Parisi et al, 2014)
When the target class image appears, the resultant P300 response provides a discernible signal for recognition by machine learning methods, allowing a BCI system to discriminate between target and non-target

Summary

Introduction

Brain-Computer Interfaces (BCIs) are augmentative devices that decode user intent directly from the brain (Wolpaw et al, 2002). Recent advances in signal processing and machine learning techniques have enabled the application of BCI technologies to fields such as medicine, industry, and recreation (Blankertz et al, 2010; Lance et al, 2012; van Erp et al, 2012). The use of advanced signal processing and machine learning techniques to minimize or eliminate this need for calibration data is an area of on-going interest for the development of practical BCI systems (Lotte, 2015). With the development of a computational framework for statistical inference (Pennec et al, 2006), information geometry has been successfully applied to radar signal processing (Barbaresco, 2008), diffusion tensor imaging (Fletcher and Joshi, 2004), and computer vision (Tuzel et al, 2007). Information geometry has been applied successfully to BCI paradigms such as motor imagery (Barachant et al, 2012, 2013) and event related potential (Barachant and Congedo, 2014)

Objectives

Methods

Results

Discussion

Conclusion