Researchers have developed CHASMBrain, a new hierarchical framework for encoding images into fMRI data. This model uses a dual-stream Mamba architecture to distinguish between global semantic information and local spatial details, mimicking the human visual cortex. Experiments on the Natural Scenes Dataset showed CHASMBrain outperformed existing methods, achieving a Pearson correlation of 0.429 and an MSE of 0.261. Further analysis indicated that the model's patch stream is linked to early visual processing, while the CLS stream provides broader semantic context to higher-order brain areas. AI
IMPACT This research offers a novel approach to understanding visual representations by linking AI models to human brain activity, potentially advancing both fields.
RANK_REASON Academic paper detailing a new model architecture and experimental results.
AI-generated summary · Google Gemini · from 3 sources. How we write summaries →
