English
Related papers

Related papers: Cautious Weight Decay

200 papers

Modern optimizers such as AdamW, equipped with momentum and adaptive learning rate, are designed to escape local minima and explore the vast parameter space. This exploration is beneficial for finding good loss basins when training from…

Machine Learning · Computer Science 2024-11-05 Junjiao Tian , Chengyue Huang , Zsolt Kira

Weight decay (WD) is a traditional regularization technique in deep learning, but despite its ubiquity, its behavior is still an area of active research. Golatkar et al. have recently shown that WD only matters at the start of the training…

Machine Learning · Computer Science 2020-12-29 Johan Bjorck , Kilian Weinberger , Carla Gomes

Adaptive optimizers like AdamW apply uniform hyperparameters across all parameter groups, ignoring heterogeneous optimization dynamics across layers and modules. We address this limitation by proposing MetaAdamW - a new optimizer that…

Machine Learning · Computer Science 2026-05-07 JiangBo Zhao , ZhaoXin Liu

Weight decay is a simple yet powerful regularization technique that has been very widely used in training of deep neural networks (DNNs). While weight decay has attracted much attention, previous studies fail to discover some overlooked…

Machine Learning · Computer Science 2024-08-19 Zeke Xie , Zhiqiang Xu , Jingzhao Zhang , Issei Sato , Masashi Sugiyama

AdamW has been the default optimizer for transformer pretraining. For many years, our community searched for faster and more stable optimizers with only constrained positive outcomes. In this work, we propose a \textbf{one-line modification…

Machine Learning · Computer Science 2026-02-17 Kaizhao Liang , Lizhang Chen , Bo Liu , Qiang Liu

Is the standard weight decay in AdamW truly optimal? Although AdamW decouples weight decay from adaptive gradient scaling, a fundamental conflict remains: the Radial Tug-of-War. In deep learning, gradients tend to increase parameter norms…

Machine Learning · Computer Science 2026-02-06 Hao Chen , Jinghui Yuan , Hanmin Zhang

Adaptive optimizers with decoupled weight decay, such as AdamW, are the de facto standard for pre-training large transformer-based generative models. Yet the quadratic nature of the $\ell_2$ penalty embedded in weight decay drives all…

Machine Learning · Computer Science 2025-11-19 Fu-Ming Guo , Yingfang Fan

L$_2$ regularization and weight decay regularization are equivalent for standard stochastic gradient descent (when rescaled by the learning rate), but as we demonstrate this is \emph{not} the case for adaptive gradient algorithms, such as…

Machine Learning · Computer Science 2019-01-08 Ilya Loshchilov , Frank Hutter

In this paper, we investigate the convergence properties of a wide class of Adam-family methods for minimizing quadratically regularized nonsmooth nonconvex optimization problems, especially in the context of training nonsmooth neural…

Optimization and Control · Mathematics 2023-10-16 Kuangyu Ding , Nachuan Xiao , Kim-Chuan Toh

Optimization algorithms with momentum, e.g., (ADAM), have been widely used for building deep learning models due to the faster convergence rates compared with stochastic gradient descent (SGD). Momentum helps accelerate SGD in the relevant…

Machine Learning · Computer Science 2020-01-24 Jiyang Bai , Yuxiang Ren , Jiawei Zhang

The vast majority of successful deep neural networks are trained using variants of stochastic gradient descent (SGD) algorithms. Recent attempts to improve SGD can be broadly categorized into two approaches: (1) adaptive learning rate…

Machine Learning · Computer Science 2019-12-04 Michael R. Zhang , James Lucas , Geoffrey Hinton , Jimmy Ba

This work advances floating-point program verification by introducing Augmented Weak-Distance (AWD), a principled extension of the Weak-Distance (WD) framework. WD is a recent approach that reformulates program analysis as a numerical…

Programming Languages · Computer Science 2025-05-21 Zhoulai Fu , Freek Verbeek , Binoy Ravindran

Decoupled weight decay, solely responsible for the performance advantage of AdamW over Adam, has long been set to proportional to learning rate $\gamma$ without questioning. Some researchers have recently challenged such assumption and…

Machine Learning · Computer Science 2026-04-15 Jason Chuan-Chih Chou

Within the context of intelligent manufacturing, industrial robots have a pivotal function. Nonetheless, extended operational periods cause a decline in their absolute positioning accuracy, preventing them from meeting high precision. To…

Robotics · Computer Science 2024-08-23 Tinghui Chen , Shuai Li

We note that decoupled weight decay regularization is a particular case of weight norm control where the target norm of weights is set to 0. Any optimization method (e.g., Adam) which uses decoupled weight decay regularization…

Machine Learning · Computer Science 2023-11-22 Ilya Loshchilov

Regularization in the optimization of deep neural networks is often critical to avoid undesirable over-fitting leading to better generalization of model. One of the most popular regularization algorithms is to impose L-2 penalty on the…

Machine Learning · Computer Science 2019-08-09 Kensuke Nakamura , Byung-Woo Hong

Introduced in the late 1980s for generalization purposes, pruning has now become a staple for compressing deep neural networks. Despite many innovations in recent decades, pruning approaches still face core issues that hinder their…

Neural and Evolutionary Computing · Computer Science 2022-03-10 Hugo Tessier , Vincent Gripon , Mathieu Léonardon , Matthieu Arzel , Thomas Hannagan , David Bertrand

Vision-language models like CLIP demonstrate impressive zero-shot generalization but remain highly vulnerable to adversarial attacks. In this work, we propose Confidence-Aware Weighting (CAW) to enhance zero-shot robustness in…

Computer Vision and Pattern Recognition · Computer Science 2025-10-06 Nikoo Naghavian , Mostafa Tavassolipour

In this paper, we introduce weight prediction into the AdamW optimizer to boost its convergence when training the deep neural network (DNN) models. In particular, ahead of each mini-batch training, we predict the future weights according to…

Machine Learning · Computer Science 2023-08-09 Lei Guan

We propose adaptive weight decay, which automatically tunes the hyper-parameter for weight decay during each training iteration. For classification problems, we propose changing the value of the weight decay hyper-parameter on the fly based…

Machine Learning · Computer Science 2023-12-05 Amin Ghiasi , Ali Shafahi , Reza Ardekani
‹ Prev 1 2 3 10 Next ›