What is it about?

This paper introduces a novel approach to estimate cognitive workload (CW), crucial in areas like healthcare, human-machine interaction, and error prediction. Traditionally, CW estimation relies on privacy-intrusive sensors gathering sensitive data like eye movement and physiological signals. Our study employs Federated Learning (FL), a privacy-centric method, to build global models without exposing individual data. We integrated local data-preprocessing and feature extraction suitable for the federated setting. Different traditional Machine Learning models, feature-based neural networks, and End-to-End models were tested on two datasets, COLET and ADABase, involving 75 participants. We also analyzed the impact of various parameters, including window length and participation rate. The results demonstrate that FL matches the accuracy of conventional methods while significantly enhancing data privacy.

Featured Image

Why is it important?

Our research is crucial as it addresses the dual challenges in CW estimation: maintaining high accuracy and ensuring user privacy. By employing Federated Learning, we developed a method that respects data privacy while obtaining comparable results to traditional CW estimation methods. We also demonstrate how data preprocessing and feature extraction can be effectively done locally on the user device. This is especially crucial in today's privacy-conscious world. Our method is distinct as it successfully applies FL to two different datasets, proving its effectiveness and adaptability in practical scenarios. This progress is significant for healthcare and other sensitive areas, where maintaining data privacy is paramount.

Perspectives

From my perspective, this research marks a major breakthrough in aligning data privacy with technological advancement in cognitive workload assessment. Addressing the challenge of protecting sensitive physiological data while accurately estimating cognitive workload has been a significant issue. The successful implementation of Federated Learning in this area not only leads to more ethical AI use in healthcare but also establishes a new norm for privacy-preserving data analysis. My future plans include expanding this research, aiming to further enhance performance by introducing client personalisation methods, and combining the two datasets in a global federated training using a multimodal model.

dario fenoglio
Universita della Svizzera Italiana

Read the Original

This page is a summary of: Federated Learning for Privacy-aware Cognitive Workload Estimation, December 2023, ACM (Association for Computing Machinery),
DOI: 10.1145/3626705.3627783.
You can read the full text:

Read

Contributors

The following have contributed to this page