Transfer Learning with Active Sampling for Rapid Training and Calibration in BCI-P300 Across Health States and Multi-centre Data

TL;DR

This work tackles the challenge of generalizing P300-BCI models amidst high inter-subject, inter-session, and inter-site variability by introducing Adaptive Transfer Learning with Dense Poison Disk Sampling (Dense PDS) as Active Sampling (AS). The method strategically samples source and target-domain data to accelerate calibration and reduce overfitting, enabling rapid, subject-adaptive fine-tuning. On two internationally replicated datasets, the approach achieves robust mean accuracy with reduced variability and significantly lower training time, demonstrating practical generalizability for real-world BCI deployment. The findings offer a pathway to faster, more reliable P300 decoding across diverse health states and hardware, with potential applicability to other EEG-based BCM paradigms and real-time systems.

Abstract

Machine learning and deep learning advancements have boosted Brain-Computer Interface (BCI) performance, but their wide-scale applicability is limited due to factors like individual health, hardware variations, and cultural differences affecting neural data. Studies often focus on uniform single-site experiments in uniform settings, leading to high performance that may not translate well to real-world diversity. Deep learning models aim to enhance BCI classification accuracy, and transfer learning has been suggested to adapt models to individual neural patterns using a base model trained on others' data. This approach promises better generalizability and reduced overfitting, yet challenges remain in handling diverse and imbalanced datasets from different equipment, subjects, multiple centres in different countries, and both healthy and patient populations for effective model transfer and tuning. In a setting characterized by maximal heterogeneity, we proposed P300 wave detection in BCIs employing a convolutional neural network fitted with adaptive transfer learning based on Poison Sampling Disk (PDS) called Active Sampling (AS), which flexibly adjusts the transition from source data to the target domain. Our results reported for subject adaptive with 40% of adaptive fine-tuning that the averaged classification accuracy improved by 5.36% and standard deviation reduced by 12.22% using two distinct, internationally replicated datasets. These results outperformed in classification accuracy, computational time, and training efficiency, mainly due to the proposed Active Sampling (AS) method for transfer learning.

Figures (12)

• Figure 1: In the present work, we validate the efficacy of adding Active Sampling (AS) by testing the EEG-based P300 decoding task-based using adaptive fine-tuning in deep learning models for subject 1 as the target subject and subjects 2 to 17 as source subjects, which represent the OE+ME w/ AS experimental scheme. The source and target subject are sampled using the AS to diversify and reduce the non-redundant data. The subject-independent and subject-adaptive schemes are represented by different colors in this figure. They and the arrows demonstrate the different blocks considered when computing the adaptive fine-tuning in deep learning models in a P300 BCI for train and testing
• Figure 2: Scheme independent data splitting
• Figure 3: Adaptive learning schemes contrasted with network layers
• Figure 4: Adaptive scheme of the subject target
• Figure 5: Interconnection between subject independent and adaptive with inputs of Source subjects (S.S) and Target subject (T.S)

Theorems & Definitions (1)

• proof