Online Classifier Adaptation in Brain-Computer Interfaces

Brain-computer interfaces (BCIs) aim to provide a new channel of communication by enabling the subject to control an external systems by using purely mental commands. One method of doing this without invasive surgical procedures is by measuring the electrical activity of the brain on the scalp through electroencephalography (EEG). A major obstacle to developing complex EEG-based BCI systems that provide a number of intuitive mental commands is the high variability of EEG signals. EEG signals from the same subject vary considerably within a single session and between sessions on the same or different days. To deal with this we are investigating methods of adapting the classifier while it is being used by the subject. By keeping the classifier constantly tuned to the EEG signals of the current session we hope to improve the performance of the classifier and allow the subject to learn to use the BCI more effectively. This paper discusses preliminary offline and online experiments towards this goal, focusing on the initial training period when the task that the subject is trying to achieve is known and thus supervised adaptation methods can be used. In these experiments the subjects were asked to perform three mental commands (imagination of left and right hand movements, and a language task) and the EEG signals were classified with a Gaussian classifier.

Adaptation in Brain-Computer Interfaces

Anna Buttfield

One major challenge in Brain-Computer Interface (BCI) research is to cope with the inherent nonstationarity of the recorded brain signals caused by changes in the subjects brain processes during an experiment. Online adaptation of the classifier embedded into the BCI is a possible way of tackling this issue. In this chapter we investigate the effect of adaptation on the performance of the classifier embedded in three different BCI systems, all of them based on non-invasive electroencephalogram (EEG) signals. Through this adaptation we aim to keep the classifier constantly tuned to the EEG signals it is receiving in the current session. Although the experimental results reported here show the benefits of online adaptation, some questions need still to be addressed. The chapter ends discussing some of these open issues.

The MIT Press2007

Towards a Robust BCI: Error Potentials and Online Learning

Anna Buttfield, Pierre Ferrez

Recent advances in the field of Brain-Computer Interfaces (BCIs) have shown that BCIs have the potential to provide a powerful new channel of communication, completely independent of muscular and nervous systems. However, while there have been successful laboratory demonstrations, there are still issues that need to be addressed before BCIs can be used by non-experts outside the laboratory. At IDIAP we have been investigating several areas that we believe will allow us to improve the robustness, flexibility and reliability of BCIs. One area is recognition of cognitive error states, that is, identifying errors through the brain's reaction to mistakes. The production of these error potentials (ErrP) in reaction to an error made by the user is well established. We have extended this work by identifying a similar but distinct ErrP that is generated in response to an error made by the interface, (a misinterpretation of a command that the user has given). This ErrP can be satisfactorily identified in single trials and can be demonstrated to improve the theoretical performance of a BCI. A second area of research is online adaptation of the classifier. BCI signals change over time, both between sessions and within a single session, due to a number of factors. This means that a classifier trained on data from a previous session will probably not be optimal for a new session. In this paper we present preliminary results from our investigations into supervised online learning that can be applied in the initial training phase. We also discuss the future direction of this research, including the combination of these two currently separate issues to create a potentially very powerful BCI.

2006

Prior knowledge in Kernel methods

Alexei Pozdnoukhov

Machine Learning is a modern and actively developing field of computer science, devoted to extracting and estimating dependencies from empirical data. It combines such fields as statistics, optimization theory and artificial intelligence. In practical tasks, the general aim of Machine Learning is to construct algorithms able to generalize and predict in previously unseen situations based on some set of examples. Given some finite information, Machine Learning provides ways to exract knowledge, describe, explain and predict from data. Kernel Methods are one of the most successful branches of Machine Learning. They allow applying linear algorithms with well-founded properties such as generalization ability, to non-linear real-life problems. Support Vector Machine is a well-known example of a kernel method, which has found a wide range of applications in data analysis nowadays. In many practical applications, some additional prior knowledge is often available. This can be the knowledge about the data domain, invariant transformations, inner geometrical structures in data, some properties of the underlying process, etc. If used smartly, this information can provide significant improvement to any data processing algorithm. Thus, it is important to develop methods for incorporating prior knowledge into data-dependent models. The main objective of this thesis is to investigate approaches towards learning with kernel methods using prior knowledge. Invariant learning with kernel methods is considered in more details. In the first part of the thesis, kernels are developed which incorporate prior knowledge on invariant transformations. They apply when the desired transformation produce an object around every example, assuming that all points in the given object share the same class. Different types of objects, including hard geometrical objects and distributions are considered. These kernels were then applied for images classification with Support Vector Machines. Next, algorithms which specifically include prior knowledge are considered. An algorithm which linearly classifies distributions by their domain was developed. It is constructed such that it allows to apply kernels to solve non-linear tasks. Thus, it combines the discriminative power of support vector machines and the well-developed framework of generative models. It can be applied to a number of real-life tasks which include data represented as distributions. In the last part of the thesis, the use of unlabelled data as a source of prior knowledge is considered. The technique of modelling the unlabelled data with a graph is taken as a baseline from semi-supervised manifold learning. For classification problems, we use this apporach for building graph models of invariant manifolds. For regression problems, we use unlabelled data to take into account the inner geometry of the input space. To conclude, in this thesis we developed a number of approaches for incorporating some prior knowledge into kernel methods. We proposed invariant kernels for existing algorithms, developed new algorithms and adapted a technique taken from semi-supervised learning for invariant learning. In all these cases, links with related state-of-the-art approaches were investigated. Several illustrative experiments were carried out on real data on optical character recognition, face image classification, brain-computer interfaces, and a number of benchmark and synthetic datasets.

EPFL2006

Prior knowledge in Kernel methods

Alexei Pozdnoukhov

EPFL2006