Search papers, labs, and topics across Lattice.
University of Minnesota
2
0
3
Straggler-Aware Group Control can significantly enhance the efficiency of synchronous reinforcement learning by dynamically optimizing group sizes, leading to faster training and better model performance.
Muon's "one-size-fits-all" spectral whitening can cripple VLA and RL, but a high-pass spectral filter (Pion) can restore performance by suppressing gradient noise and preserving pre-trained head specialization.
