- New preprint (#neuroscience #deeplearning doi.org/10.1101/2025...)! We trained 20 DCNNs on 941235 images with varying scene segmentation (original. object-only, silhouette, background-only). Despite object recognition varying (27-53%), all networks showed similar EEG prediction.Mar 15, 2025 13:55
- Key finding: Network layers better at categorizing objects are WORSE at predicting brain activity. EEG signals reflect scene segmentation, not object recognition. This explains why object recognition models don't better predict neural responses. #ComputationalNeuroscience #AI
- Surprising result: Networks trained on background-only images (lowest categorization: 27%) predicted brain activity as well as those trained on complete scenes. Early visual cortex prioritizes scene segmentation over object recognition! #neuroimaging #computervision #EEG
- Implications: To better align AI with human vision, focus on scene segmentation, not just object recognition. Our findings suggest biological vision fundamentally differs from current DCNNs in how it processes complex scenes. Paper: doi.org/10.1101/2025... #AI #neuroscience
