Combining General and Personalized Models for Epilepsy Detection with Hyperdimensional Computing

TL;DR

This work investigates hyperdimensional computing (HD) as a low-power framework for epilepsy detection in wearables, focusing on personalized, generalized, and hybrid models. It demonstrates how HD representations enable direct inter-subject comparisons, derives generalized models from personalized ones using averaging and weighted schemes, and shows that hybrid models combining both paradigms can improve detection performance. The study also explores knowledge transfer across EEG datasets, revealing direction-dependent gains and practical strategies for reusing models with selective retraining. Overall, the results highlight HD's potential for wearables to deliver accurate, privacy-preserving seizure monitoring and to provide neurological insights into individual patterns, with limitations tied to dataset heterogeneity and seizure-type variability.

Abstract

Epilepsy is a chronic neurological disorder with a significant prevalence. However, there is still no adequate technological support to enable epilepsy detection and continuous outpatient monitoring in everyday life. Hyperdimensional (HD) computing is an interesting alternative for wearable devices, characterized by a much simpler learning process and also lower memory requirements. In this work, we demonstrate a few additional aspects in which HD computing, and the way its models are built and stored, can be used for further understanding, comparing, and creating more advanced machine learning models for epilepsy detection. These possibilities are not feasible with other state-of-the-art models, such as random forests or neural networks. We compare inter-subject similarity of models per different classes (seizure and non-seizure), then study the process of creation of generalized models from personalized ones, and in the end, how to combine personalized and generalized models to create hybrid models. This results in improved epilepsy detection performance. We also tested knowledge transfer between models created on two different datasets. Finally, all those examples could be highly interesting not only from an engineering perspective to create better models for wearables, but also from a neurological perspective to better understand individual epilepsy patterns.

Figures (10)

• Figure 1: HD workflow for training classical and online HD models. Online training differs in that the class vectors are updated after every datapoint by multiplying its similarity to the target class by the vector before accumulating it into the class.
• Figure 2: Statistics of the Repomse dataset: number of seizures per patient in original database, and also reasons for rejecting specific subjects. In the end we keep 286 subjects which satisfy our criteria.
• Figure 3: Inter-subject similarity between model vectors of individual subjects (both between ictal (Seiz, S) or inter-ictal (NonSeiz, NS) models).
• Figure 4: Comparing different approaches to creating generalized models from personalized. Similarity between seizure and non-seizure model vectors as well as overall class separability is measured.
• Figure 5: Evolution of generalized vectors as adding one by one individual subject. Average similarity with all personalized subjects is measured to characterize how stable are generalized vectors.