Machine learning methods in assistive technologies

Assistive devices are essential in enhancing the quality of life for individuals who have severe disabilities, such as quadriplegia and amyotrophic lateral sclerosis, or who have had massive brainstem strokes. However, the effectiveness of these systems is dependent on preserved residual movements or speech. In the absence of means to repair the damaged nervous system, three options exist for restoring function: 1) augmenting the capabilities of remaining pathways, 2) detouring around points of damage, or 3) providing the brain with new channels for communication and control. The paper reviews the use of machine learning methods for development of assistive technology. Three projects are described, representing the three options listed above. In each of them machine learning methods are employed to help with pattern recognition and classification. The three projects are: automatic speech recognition of dysarthric speech; control strategies for FES-assisted locomotion (functional electrical stimulation); and an EEG-based computer access. Although these three projects may look very different from each other, the structure of their experimental set-ups, and their potential for application in assistive devices are very similar. All experimental set-ups consist of sensory signal acquisition, signal processing for feature extraction, and data processing by machine learning techniques for pattern recognition and classification. In addition, all three projects deal with digital signal processing and machine learning method applications in development of man-machine interfaces.