EEG-Reptile: An Automatized Reptile-Based Meta-Learning Library for BCIs

TL;DR

The paper addresses the challenge of inter-subject variability and limited per-user data in EEG-based BCIs by introducing EEG-Reptile, an automated library that applies the Reptile meta-learning algorithm to rapidly adapt neural networks to new subjects. It integrates automated hyperparameter tuning, data management, and support for multiple architectures (e.g., EEGNet, EEG-Inception, FBCNet), enabling zero-shot and few-shot learning in MI-BCI tasks. The authors demonstrate improvements over baseline transfer learning on two MI datasets (BCI IV 2a and Lee2019 MI), with notable gains when using meta-learning and optimized network configurations, and they introduce an outlier-filtering weight initialization to improve inter-subject transfer. The work offers a practical, scalable tool for researchers and practitioners to automate meta-learning for EEG, potentially reducing calibration time and expanding the applicability of BCI systems in neurorehabilitation and related domains.

Abstract

Meta-learning, i.e., "learning to learn", is a promising approach to enable efficient BCI classifier training with limited amounts of data. It can effectively use collections of in some way similar classification tasks, with rapid adaptation to new tasks where only minimal data are available. However, applying meta-learning to existing classifiers and BCI tasks requires significant effort. To address this issue, we propose EEG-Reptile, an automated library that leverages meta-learning to improve classification accuracy of neural networks in BCIs and other EEG-based applications. It utilizes the Reptile meta-learning algorithm to adapt neural network classifiers of EEG data to the inter-subject domain, allowing for more efficient fine-tuning for a new subject on a small amount of data. The proposed library incorporates an automated hyperparameter tuning module, a data management pipeline, and an implementation of the Reptile meta-learning algorithm. EEG-Reptile automation level allows using it without deep understanding of meta-learning. We demonstrate the effectiveness of EEG-Reptile on two benchmark datasets (BCI IV 2a, Lee2019 MI) and three neural network architectures (EEGNet, FBCNet, EEG-Inception). Our library achieved improvement in both zero-shot and few-shot learning scenarios compared to traditional transfer learning approaches.

Figures (5)

• Figure 1: Structure of EEG-Reptile library.
• Figure 2: Experiment design.
• Figure 3: Comparison of group average Accuracy with 95% confidence intervals on BCI IV 2a (four classes) and Lee2019 (MI) (two classes) datasets for baseline algorithms and algorithms with meta-learning fine-tuned on subsets with different sizes and zero-shot.
• Figure 4: Comparison of group average of difference for Accuracy with 95% confidence intervals on BCI IV 2a (four classes) and Lee2019 (MI) (two classes) datasets between baseline algorithms and algorithms with meta-learning fine-tuned on subsets with different sizes and zero-shot.
• Figure 5: Comparison of group average Accuracy with 95% confidence intervals on BCI IV 2a (four classes) dataset between baseline algorithms and algorithms with meta-learning applied with EEGNet optimization and without, fine-tuned on subsets with different sizes and zero-shot.