Talks

K-POP Talk

K-POP Optimizer with Tycho van der OuderraAugust 2025

Cohere Reading Group

Talk on K-POP, a novel optimizer applying Adam in the Kronecker-factored eigenbasis.

Talk Link
SPARTA Talk

SPARTA: Sparse Parameter AveragingMarch 2025

ICLR 2025 MCDC Workshop

Talk on SPARTA, a novel distributed training method with reduced communication overhead.

Talk Link