Deep Learning Architectures Used In Eeg-Based Estimation Of Cognitive workload: A Review

Cognitive workload (CWL) refers to the ratio of a participant’s mental effort over his/her brain capacity when executing tasks with aid of a machine. Such CWL influences the participant’s trust placed on the machine and thus affects the tasks’ performance. Efficient human-machine interaction demands the machine’s real-time adaptation to meet an admissible CWL for the participant. The adaptation needs estimating CWL based on brain activities captured by non-invasive electroencephalography (EEG). Since deep learning (DL) is common for extracting EEG features reflecting certain characteristics of the activities, DL-based CWL estimation attracts ample attention. Herein, we present a review to summarize current trends in DL architectures for EEG-based CWL estimation and to identify gaps in the trends for future work.

1. Introduction

Cognitive workload (CWL) is defined as a quantitative measure of mental efforts forced on cognitive resources (e.g., working memory) of the human brain while performing a task [1]. As the brain resources are constrained, tasks overloading cognition may reduce efficiency and result in critical errors. In contrast, a deficient workload is a waste of the resources and leads to boredom during task execution [2]. In addition, CWL is inversely proportional to trust – an intrinsic characteristic needed for an autonomous human-machine system (HMS) [3], [4]. An HMS requires collaboration between a participant and a machine to bring forth flexible decisionmaking [5]. The collaboration would ideally be constructive if the participant could trust the machine to undertake designated actions. The trust needs the machine to be adaptive for ensuring the participant’s admissible CWL. Estimating the CWL is thus crucial to assure the performance of the HMS.