Through controlled model-to-brain comparisons across a large-scale survey of deep neural networks, the authors show the data models are trained on matters far more for downstream brain prediction than design factors such as architecture and training task.
- Colin Conwell
- Jacob S. Prince
- Talia Konkle
