Zhejiang University Research Team Develops AI Model Mimicking Human Brain Conceptual Understanding
A research team from Zhejiang University has published a paper in Nature Communications, proposing a novel approach to AI model development that mimics human brain conceptual understanding. The study challenges the mainstream view that increasing model parameters leads to human-like thinking. Instead, the team found that while larger models improve specific object recognition, they struggle with abstract concept understanding. The research involved using brain signal data to supervise AI models, aiming to transfer human conceptual structures to deep neural networks (DNNs). Experiments showed that models trained with brain signals performed better in novel situations, improving abstract concept recognition by 20.5% compared to control models with more parameters. 
