Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
Email:
1
0
2
The loss surface of a simple polynomial activation network exactly solves its own training dynamics, revealing a blueprint for training hundred-layer networks without normalization.