Academic journal article Human Factors

Monitoring Working Memory Load during Computer-Based Tasks with EEG Pattern Recognition Methods

Academic journal article Human Factors

Monitoring Working Memory Load during Computer-Based Tasks with EEG Pattern Recognition Methods

Article excerpt

INTRODUCTION

In this paper we document the evaluation of a method for deriving a continuous index of the working memory load that is incurred by individuals who are engaged in computer-based tasks. Working memory refers to the limited capacity for holding information in mind for several seconds in the context of cognitive activity (Baddeley & Hitch, 1974). Overload of working memory has long been recognized as an important source of performance errors during human-computer interaction (e.g., Card, Moran, & Newell, 1983; Kieras, 1988; Olson & Olson, 1990). This source of error is particularly acute in unskilled users, for whom unfamiliar procedures are likely to require greater commitment of cognitive resources (Anderson & Boyle, 1987; Carlson, Sullivan, & Schneider, 1989). Furthermore, overload of working memory capacity has been found to be a limiting factor in the early stages of procedural skill acquisition (Kyllonen & Stephens, 1990; Woltz, 1988). As a result, the need to minimize working memory load has been cited as a primary guiding principle for the design of intelligent tutoring systems (e.g., Anderson & Boyle, 1987). An effective means of monitoring the working memory demands of computer-based tasks would therefore be a useful tool for evaluating alternative interface designs and computer-based training protocols.

Applied contexts require that a candidate method for monitoring working memory load must not interfere with operator performance, must be employable across many contexts, and must have reasonably good time resolution. Furthermore, such a method must be robust enough to be reliably measured under relatively unstructured task conditions yet sensitive enough to consistently vary with some dimension of interest. Some physiological measurements, particularly measures of central nervous system activity such as the electroencephalogram (EEG), appear to meet these requirements. Indeed, EEG measures have frequently been shown to be highly sensitive to variations in task difficulty (Gevins, Smith, McEvoy, & Yu, 1997; Gundel & Wilson, 1992; Humphrey & Kramer, 1994; Isreal, Wickens, Chesney, & Donchin, 1980; Parasuraman, 1990; Sterman, 1989; Sterman, Mann, Kaiser, & Suyenobu, 1994). In addition, multivariate combinations of EEG variables can be used to accurately discriminate between specific cognitive states (e.g., Gevins, Zeitlin, Doyle, Schaffer, & Callaway, 1979; Gevins, Zeitlin, Doyle, Yinglin, et al., 1979; Gevins, Zeitlin, Yingling, et al., 1979; Wilson & Fisher, 1995).

Preliminary evidence that such measures can be used to automate recognition of patterns of neural activity associated with high working memory loads was obtained in a pilot study. Specifically, Leong and Gevins (1994) used multivariate combinations of EEG features to train neural networks to discriminate a cognitive state associated with performance of a high-load working memory task from a low-load control condition. An accuracy of over 90% correct classification of data segments was achieved on test data sets from four individual participants. The present experiment was designed to replicate those results in a larger number of participants and to extend the technique over a wider range of load levels and task conditions. Specifically, the analysis was extended to identify intermediate working memory load levels, to examine the test-retest reliability of EEG-based pattern recognition networks, and to examine how classification networks could be generalized across tasks and participants.

METHOD

Participants

Eight healthy volunteers (five men and three women) 22 to 28 years of age (mean = 24 years) participated in the study and were paid on an hourly basis.

Task Structure and Procedure

Participants performed a continuous matching task that required them to indicate whether the current stimulus matched a stimulus presented on a previous trial. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.