Subject Adaptive EEG-based Visual Recognition (2110.13470v1)

Abstract: This paper focuses on EEG-based visual recognition, aiming to predict the visual object class observed by a subject based on his/her EEG signals. One of the main challenges is the large variation between signals from different subjects. It limits recognition systems to work only for the subjects involved in model training, which is undesirable for real-world scenarios where new subjects are frequently added. This limitation can be alleviated by collecting a large amount of data for each new user, yet it is costly and sometimes infeasible. To make the task more practical, we introduce a novel problem setting, namely subject adaptive EEG-based visual recognition. In this setting, a bunch of pre-recorded data of existing users (source) is available, while only a little training data from a new user (target) are provided. At inference time, the model is evaluated solely on the signals from the target user. This setting is challenging, especially because training samples from source subjects may not be helpful when evaluating the model on the data from the target subject. To tackle the new problem, we design a simple yet effective baseline that minimizes the discrepancy between feature distributions from different subjects, which allows the model to extract subject-independent features. Consequently, our model can learn the common knowledge shared among subjects, thereby significantly improving the recognition performance for the target subject. In the experiments, we demonstrate the effectiveness of our method under various settings. Our code is available at https://github.com/DeepBCI/Deep-BCI/tree/master/1_Intelligent_BCI/Subject_Adaptive_EEG_based_Visual_Recognition.

• The paper presents a subject-adaptive EEG framework that minimizes inter-subject variability in visual recognition tasks.
• It employs maximum mean discrepancy (MMD) to extract subject-independent features, achieving a 6.4% gain in extreme 1-shot settings.
• The approach reduces costly data collection and enables rapid adaptation for practical brain-computer interface applications.

Subject Adaptive EEG-based Visual Recognition

The paper "Subject Adaptive EEG-based Visual Recognition" presents a novel approach to improving the robustness and applicability of EEG-based visual recognition systems. The research addresses the challenges posed by inter-subject variability in EEG signals, which have traditionally restricted recognition systems to subjects involved in the model training phase. This limitation is particularly problematic in real-world applications where new subjects are frequently added, necessitating costly and time-consuming data collection.

Introduction to EEG-Based Visual Recognition

The paper begins by outlining the significance of brain-computer interfaces (BCIs) and the role of EEG in analyzing human brain activity. EEG-based models have become prevalent in various applications, such as disorder detection and emotion recognition, due to their non-invasive nature and rapid data acquisition capabilities. The focus of this research is to classify visual stimuli based on EEG signals, leveraging deep learning to enhance recognition performance.

Novel Problem Setting

A key contribution of this work is the introduction of a subject-adaptive EEG-based visual recognition framework. Unlike traditional models constrained by data from known subjects, this setting allows for the integration of abundant EEG data from source subjects while only requiring minimal data from new target subjects. This approach significantly alleviates the need for extensive data collection, making it more practical and feasible for real-world applications.

Methodology

The research proposes a baseline method designed to minimize the feature distribution discrepancy across different subjects. The authors implement maximum mean discrepancy (MMD) to train models that extract subject-independent features, enabling the transfer of learned knowledge from source subjects to the target subject. This methodology effectively enhances the recognition performance, even when the data from the target subject is sparse.

Experimental Results

The experiments demonstrate the efficacy of the proposed model under various conditions. Notably, the model achieves a remarkable improvement in recognition accuracy, with a performance gain of 6.4% in an extreme 1-shot setting. This indicates the model's capability to generalize learned features across subjects, setting a strong precedent for future research in reducing training data requirements.

Implications and Future Directions

This paper's contributions have both practical and theoretical implications. Practically, the approach could be deployed in applications where rapid adaptation to new users is crucial, such as assistive technologies and personalized brain-computer interfaces. Theoretically, this work opens avenues for further research in subject adaptation and domain generalization within the field of neural data processing.

Future research could explore enhancing the model's adaptability and efficiency, perhaps integrating advanced neural architectures or auxiliary information to further bridge the subject variability gap. Additionally, investigating the application of this framework to other modalities of brain signals or multi-modal recognition tasks could extend its utility.

Conclusion

In summary, the paper delivers a significant step forward in EEG-based visual recognition, addressing a critical barrier to its broader adoption. The subject adaptive setting and the methodology for minimizing inter-subject variability present a compelling advancement, with promising potential applications in real-world scenarios where data availability is often a bottleneck.