paper-with-me

홈 › Papers

How to decay your learning rate

2021-03-23 · Aitor Lewkowycz

Complex learning rate schedules have become an integral part of deep learning. We find empirically that common fine-tuned schedules decay the learning rate after the weight norm bounces. This leads to the proposal of ABEL: an automatic scheduler which decays the learning rate by keeping track of the weight norm. ABEL's performance matches that of tuned schedules and is more robust with respect to its parameters. Through extensive experiments in vision, NLP, and RL, we show that if the weight norm does not bounce, we can simplify schedules even further with no loss in performance. In such cases, a complex schedule has similar performance to a constant learning rate with a decay at the end of training.

📄 PDF Abstract BibTeX arXiv:2103.12682

Code (1)

tourdeml/abel-pytorch pytorch

Similar Papers 제목 키워드 기반

SGD and Weight Decay Secretly Minimize the Rank of Your Neural Network

2022-06-12 · Tomer Galanti, Zachary S. Siegel, Aparna Gupte, Tomaso Poggio

We investigate the inherent bias of Stochastic Gradient Descent (SGD) toward learning low-rank weight matrices during the training of deep neural networks. Our results demonstrate that training with mini-batch SGD and we…

How Learning Rate Decay Wastes Your Best Data in Curriculum-Based LLM Pretraining

2025-11-24 · Kairong Luo, Zhenbo Sun, Haodong Wen, Xinyu Shi 외 arxiv

Due to the scarcity of high-quality data, large language models (LLMs) are often trained on mixtures of data with varying quality levels, even after sophisticated data curation. A natural approach to better leverage high…

Celebrating Three Decades of Worldwide Stock Market Manipulation

2019-11-20 · Bruce Knuteson

As the decade turns, we reflect on nearly thirty years of successful manipulation of the world's public equity markets. This reflection highlights a few of the key enabling ingredients and lessons learned along the way. …

Misinformation

Can Fine-Tuning Erase Your Edits? On the Fragile Coexistence of Knowledge Editing and Adaptation

2025-11-08 · Yinjie Cheng, Paul Youssef, Christin Seifert, Jörg Schlötterer 외 arxiv

Knowledge editing (KE) offers a lightweight alternative to retraining for updating large language models (LLMs). Meanwhile, fine-tuning remains the default operation for adapting LLMs to new domains and tasks. Despite th…

knowledge editing

Your Teacher Can't Help You Here: Combating Supervision Fidelity Decay in On-Policy Distillation

2026-05-29 · Yanjiang Liu, Jie Lou, Xinyan Guan, Yuqiu Ji 외 arxiv

On-policy distillation transfers reasoning capabilities by training a student model on its own generated trajectories using token-level feedback from a teacher. However, we identify a critical bottleneck, \textbf{Supervi…

Computational Efficiency