Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

ViEEG: Hierarchical Visual Neural Representation for EEG Brain Decoding

Created by
  • Haebom

Author

Minxu Liu, Donghai Guan, Chuhang Zheng, Chunwei Tian, Jie Wen, Qi Zhu

Outline

This paper addresses the problem of interpreting brain activity as visual representations. We highlight the limitations of existing EEG visual decoding methods due to the Hierarchical Neural Encoding Neglect (HNEN) problem, and propose a novel framework, ViEEG, inspired by the hierarchical structure of the visual cortex. ViEEG decomposes visual stimuli into three biologically aligned components—contours, foreground objects, and background scenes—and utilizes a triple-stream EEG encoder based on these components. Cross-attention routing mimics the flow of low-level to high-level visual information, and hierarchical contrastive learning performs EEG-CLIP representation alignment to enable zero-shot object recognition. Experimental results on the THINGS-EEG and THINGS-MEG datasets demonstrate significantly superior performance to existing methods, suggesting a new paradigm for EEG brain decoding.

Takeaways, Limitations

•
Takeaways:
◦
We present a novel EEG visual decoding framework (ViEEG) that addresses the hierarchical neural encoding neglect (HNEN) problem.
◦
Improved performance by mimicking biological visual processing.
◦
Zero-shot object recognition possible.
◦
Demonstrated superior performance compared to existing methods on the THINGS-EEG and THINGS-MEG datasets.
◦
Presenting a new paradigm in the field of EEG brain decoding.
•
Limitations:
◦
Although not explicitly mentioned in the paper, future research is needed to verify whether expanding the diversity of the dataset and improving the generalization performance of the algorithm can improve its performance. Furthermore, further research may be necessary to apply this to real-world applications.
👍
Made with Slashpage