EEGMobile: Enhancing Speed and Accuracy in EEG-Based Gaze Prediction with Advanced Mobile Architectures
Teng Liang, Andrews Damoah
TL;DR
This work tackles real-time EEG-based gaze estimation on resource-limited devices by introducing EEGMobile, a lightweight model that fuses a pre-trained MobileViT with knowledge distillation from a high-capacity EEGViT-TCN teacher. The approach blends a Temporal Convolutional Network, feature extraction layers, and MobileViT to predict XY gaze coordinates on the EEGEyeNet Absolute Position task, achieving RMSE of 53.6 mm and offering about 33% faster inference with roughly 60% fewer parameters compared to ViT-based baselines. Although its RMSE is about 3% higher than the strongest SOTA ViT model, EEGMobile remains competitive while delivering substantial gains in speed and compactness, highlighting the viability of on-device EEG regression. These results have practical implications for on-device brain-computer interfaces and HCI applications in VR/AR and medical contexts, where low latency and limited hardware are critical.
Abstract
Electroencephalography (EEG) analysis is an important domain in the realm of Brain-Computer Interface (BCI) research. To ensure BCI devices are capable of providing practical applications in the real world, brain signal processing techniques must be fast, accurate, and resource-conscious to deliver low-latency neural analytics. This study presents a model that leverages a pre-trained MobileViT alongside Knowledge Distillation (KD) for EEG regression tasks. Our results showcase that this model is capable of performing at a level comparable (only 3% lower) to the previous State-Of-The-Art (SOTA) on the EEGEyeNet Absolute Position Task while being 33% faster and 60% smaller. Our research presents a cost-effective model applicable to resource-constrained devices and contributes to expanding future research on lightweight, mobile-friendly models for EEG regression.