6 papers
Adaptive Batch Sizes Using Non-Euclidean Gradient Noise Scales for Stochastic Sign and Spectral Descent
Hiroki Naganuma, Shagun Gupta, Youssef Briki +4
To maximize hardware utilization, modern machine learning systems typically employ large constant or manually tuned batch size schedules, relying on heuristics that are brittle and…
Smoothing DiLoCo with Primal Averaging for Faster Training of LLMs
Aaron Defazio, Konstantin Mishchenko, Parameswaran Raman +2
We propose Generalized Primal Averaging (GPA), an extension of Nesterov's method that unifies and generalizes recent averaging-based optimizers like single-worker DiLoCo and Schedu…
EMC: Efficient MCMC Negative Sampling for Contrastive Learning with Global Convergence
Chung-Yiu Yau, Hoi-To Wai, Parameswaran Raman +2
A key challenge in contrastive learning is to generate negative samples from a large sample set to contrast with positive samples, for learning better encoding of the data. These n…
Krylov Cubic Regularized Newton: A Subspace Second-Order Method with Dimension-Free Convergence Rate
Ruichen Jiang, Parameswaran Raman, Shoham Sabach +3
Second-order optimization methods, such as cubic regularized Newton methods, are known for their rapid convergence rates; nevertheless, they become impractical in high-dimensional…
MADA: Meta-Adaptive Optimizers through hyper-gradient Descent
Kaan Ozkara, Can Karakus, Parameswaran Raman +4
Following the introduction of Adam, several novel adaptive optimizers for deep learning have been proposed. These optimizers typically excel in some tasks but may not outperform Ad…
Contractive error feedback for gradient compression
Bingcong Li, Shuai Zheng, Parameswaran Raman +2
On-device memory concerns in distributed deep learning have become severe due to (i) the growth of model size in multi-GPU training, and (ii) the wide adoption of deep neural netwo…