Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
New York University
1
0
2
Forget hyperparameter tuning: periodically restarting outer momentum in distributed training can widen the stable range of learning rates and momentum, making training more robust.