CD-GraB: Coordinating Distributed Example Orders for Provably Accelerated Training
A. Feder Cooper · Wentao Guo · Duc Khiem Pham · Tiancheng Yuan · Charlie Ruan · Yucheng Lu · Christopher De Sa
Great Hall & Hall B1+B2 (level 1) #1027
Recent research on online Gradient Balancing (GraB) has revealed that there exist permutation-based example orderings that are guaranteed to outperform random reshuffling (RR). Whereas RR arbitrarily permutes training examples, GraB leverages stale gradients from prior epochs to order examples -- achieving a provably faster convergence rate than RR. However, GraB is limited by design: While it demonstrates an impressive ability to scale-up training on centralized data, it does not naturally extend to modern distributed ML workloads. We therefore propose Coordinated Distributed GraB (CD-GraB), which uses insights from prior work on kernel thinning to translate the benefits of provably faster permutation-based example ordering to distributed settings. With negligible overhead, CD-GraB exhibits a linear speedup in convergence rate over centralized GraB and outperforms baselines empirically, including distributed RR, on a variety of benchmark tasks.