The study of how matrix-preconditioned methods and a momentum-scheduled method scale relative to AdamW establish training horizon as an essential axis for optimizer evaluation and design.
Abstract
We investigate how optimizers scale across the overtraining axis and show that relative optimizer performance and optimal hyperparameters change substantially with training horizon. In particular, we study how matrix-preconditioned methods (Muon and SOAP) and a momentum-scheduled method (ADANA) scale relative to AdamW. We compare these four optimizers across models from 51M to 253M parameters and overtraining (OT) factors from 1x to 256x, sweeping the base learning rate at every setting. The preferred learning rate schedule can reverse across the overtraining axis, the best weight decay coefficient scales approximately as sqrt(OT), and longer horizons generally favor longer fixed memory. ADANA's scaling advantage over AdamW persists after tuning AdamW's fixed memory separately at each horizon. Log-time weight decay and momentum cooldown provide substantial gains for ADANA that compound as training increases. With this treatment, ADANA outscales AdamW with an exponent advantage close to that predicted by DANA theory on power-law random features. Muon and SOAP instead provide roughly constant token-efficiency advantages over AdamW across most of the measured range, although SOAP may gain further at the highest overtraining factors. ADANA begins behind both matrix-preconditioned optimizers but closes the gaps as training increases, surpassing Muon and becoming competitive with SOAP at our highest OT factors. These results establish training horizon as an essential axis for optimizer evaluation and design.
A plethora of new adaptive optimizers are designed to efficiently estimate and use minibatch gradient statistics to shape parameter updates, but they are typically benchmarked at a single batch size. Hyperparameter scaling rules promise to preserve performance as batch size and gradient noise change, suggesting that th...
Spectral-Aware Muon is introduced, which holds the head at the Muon scale and amplifies the bulk using a static spectral prior, and both variants outperform tuned AdamW and Muon (Scion implementation) baselines in all evaluated model-scale and batch-size configurations.
Xiaodong Wu, Wen-Yi Yu, Chao Zhang et al.· 1 citation
Evaluating deep learning optimizer is challenging because confounding factors such as hyperparameter tuning and optimizer-model co-design often obscure true algorithmic performance. We present results of the updated AlgoPerf V1 benchmark, in which we evaluate a new set of training algorithms on wall-clock time-to-targe...
Priya Kasimbeg, Ahmed Khaled, Aahlad Chandrabhatta et al.· 0 citations
Optimizer momentum is usually stored as a parameter-sized moving average of past gradients, which makes history costly and fixes each past signal in the coordinates in which it was computed. We introduce Backpropagated Output Momentum (BOM), which instead stores a compact moving average of prediction errors at the mode...
Yu-Chen Li, Zong-Qi Fan, Nguyen H. Tran et al.· 0 citations
This survey organizes recent optimizers and training optimization methods along four largely independent axes: temporal estimation, update geometry, horizon management, and representation and systems.
Periodic Row-wise Muon is introduced, which performs a full NS5 spectral update once every \(K\) steps and applies a low compute and communication cost row-wise constrained update based on the current momentum at the remaining steps to preserve Muon's generative quality advantage while translating it into end-to-end tr...
Chenghao Li, Xiao Han, Xin-Xin Huang et al.· 0 citations
Related blog posts
MIT News · Artificial Intelligence· news.mit.eduNov 17, 2026
By studying the computational foundations of language learning, Associate Professor Jacob Andreas builds intelligent systems that can communicate effectively with humans.
Microsoft Research Blog· microsoft.comNov 16, 2026
AI coding agents are increasingly help developers fix bugs and build features largely on their own, but every interaction with their models consumes tokens. In each session, an agent sends instructions, tool descriptions, and session history to its model. Most efforts to reduce this cost focus on individual tasks. Yet developers return to the same […] The post TokenEconomizer: Learning from Experience to Make AI Coding Agents More Efficient appeared first on Microsoft Research.
MIT News · Artificial Intelligence· news.mit.eduNov 13, 2026