Unlock: Mean Field Theory

The mean field limit of neural networks: as width goes to infinity under the right scaling, neurons become independent particles whose weight distribution evolves by Wasserstein gradient flow, capturing feature learning that the NTK regime misses.