English
Related papers

Related papers: Adaptive Preconditioners Trigger Loss Spikes in Ad…

200 papers

Loss spikes remain a persistent obstacle in large-scale language model pretraining. While previous research has attempted to identify the root cause of loss spikes by investigating individual factors, we observe that, in practice, such…

Machine Learning · Computer Science 2026-02-24 Guoxia Wang , Shuai Li , Congliang Chen , Jinle Zeng , Jiabin Yang , Dianhai Yu , Yanjun Ma , Li Shen

Very little is known about the training dynamics of adaptive gradient methods like Adam in deep learning. In this paper, we shed light on the behavior of these algorithms in the full-batch and sufficiently large batch settings.…

Modern adaptive optimization methods, such as Adam and its variants, have emerged as the most widely used tools in deep learning over recent years. These algorithms offer automatic mechanisms for dynamically adjusting the update step based…

Machine Learning · Computer Science 2025-02-12 Son Nguyen , Bo Liu , Lizhang Chen , Qiang Liu

Adaptive methods such as Adam and RMSProp are widely used in deep learning but are not well understood. In this paper, we seek a crisp, clean and precise characterization of their behavior in nonconvex settings. To this end, we first…

Machine Learning · Computer Science 2020-02-04 Matthew Staib , Sashank J. Reddi , Satyen Kale , Sanjiv Kumar , Suvrit Sra

Adaptive gradient optimization methods, such as Adam, are prevalent in training deep neural networks across diverse machine learning tasks due to their ability to achieve faster convergence. However, these methods often suffer from…

Machine Learning · Computer Science 2025-02-12 Abulikemu Abuduweili , Changliu Liu

Adaptive optimizers, such as Adam, have achieved remarkable success in deep learning. A key component of these optimizers is the so-called preconditioning matrix, providing enhanced gradient information and regulating the step size of each…

Machine Learning · Computer Science 2024-12-10 Yun Yue , Zhiling Ye , Jiadi Jiang , Yongchao Liu , Ke Zhang

Adaptive gradient methods, e.g. \textsc{Adam}, have achieved tremendous success in machine learning. Scaling the learning rate element-wisely by a certain form of second moment estimate of gradients, such methods are able to attain rapid…

Machine Learning · Computer Science 2022-02-10 Yizhou Wang , Yue Kang , Can Qin , Huan Wang , Yi Xu , Yulun Zhang , Yun Fu

We provide a theoretical analysis of Adam under non-stationary stochastic objectives, separating two regimes: Euclidean tracking under adaptive strong monotonicity of the Adam-preconditioned mean-gradient operator, and high-probability…

Machine Learning · Statistics 2026-05-07 Sharan Sahu , Abir Sarkar , Cameron J. Hogan , Martin T. Wells

In this work, we investigate the mechanism underlying loss spikes observed during neural network training. When the training enters a region with a lower-loss-as-sharper (LLAS) structure, the training becomes unstable, and the loss…

Machine Learning · Computer Science 2024-10-08 Xiaolong Li , Zhi-Qin John Xu , Zhongwang Zhang

Loss spikes often occur during pre-training of large language models. The spikes degrade the performance of large language models and sometimes ruin the pre-training. Since the pre-training needs a vast computational budget, we should avoid…

Computation and Language · Computer Science 2025-07-28 Sho Takase , Shun Kiyono , Sosuke Kobayashi , Jun Suzuki

The dynamic behavior of RMSprop and Adam algorithms is studied through a combination of careful numerical experiments and theoretical explanations. Three types of qualitative features are observed in the training loss curve: fast initial…

Machine Learning · Computer Science 2021-10-01 Chao Ma , Lei Wu , Weinan E

We observe that the traditional use of DP with the Adam optimizer introduces a bias in the second moment estimation, due to the addition of independent noise in the gradient computation. This bias leads to a different scaling for low…

Machine Learning · Computer Science 2023-04-25 Qiaoyue Tang , Mathias Lécuyer

It is widely believed that stochastic gradient descent (SGD) performs significantly worse than adaptive optimizers such as Adam in pre-training Large Language Models (LLMs). Yet the underlying reason for this gap remains unclear. In this…

Machine Learning · Computer Science 2026-05-19 Athanasios Glentis , Dawei Li , Chung-Yiu Yau , Mingyi Hong

In this paper, we provide a rigorous proof of convergence of the Adaptive Moment Estimate (Adam) algorithm for a wide class of optimization objectives. Despite the popularity and efficiency of the Adam algorithm in training deep neural…

Optimization and Control · Mathematics 2023-11-08 Haochuan Li , Alexander Rakhlin , Ali Jadbabaie

Despite the popularity of the Adam optimizer in practice, most theoretical analyses study Stochastic Gradient Descent (SGD) as a proxy for Adam, and little is known about how the solutions found by Adam differ. In this paper, we show that…

Machine Learning · Computer Science 2025-11-05 Xinghan Li , Haodong Wen , Kaifeng Lyu

The adaptive optimizer for training neural networks has continually evolved to overcome the limitations of the previously proposed adaptive methods. Recent studies have found the rare counterexamples that Adam cannot converge to the optimal…

Machine Learning · Computer Science 2019-11-04 Kiwook Bae , Heechang Ryu , Hayong Shin

Gradient descent prevails in artificial neural network training, but seems inept for spiking neural networks as small parameter changes can cause sudden, disruptive (dis-)appearances of spikes. Here, we demonstrate exact gradient descent…

Neurons and Cognition · Quantitative Biology 2025-01-29 Christian Klos , Raoul-Martin Memmesheimer

Training large language models (LLMs) presents numerous challenges, including gradient instability and loss spikes. These phenomena can lead to catastrophic divergence, requiring costly checkpoint restoration and data batch skipping.…

Machine Learning · Computer Science 2025-04-04 Abhay Kumar , Louis Owen , Nilabhra Roy Chowdhury , Fabian Güra

Many continual-learning methods modify gradients upstream (e.g., projection, penalty rescaling, replay mixing) while treating Adam as a neutral backend. We show this composition has a hidden failure mode. In a high-overlap, non-adaptive…

Machine Learning · Computer Science 2026-04-27 Yuelin Hu , Zhenbo Yu , Zhengxue Cheng , Wei Liu , Li Song

Adaptive optimization methods are well known to achieve superior convergence relative to vanilla gradient methods. The traditional viewpoint in optimization, particularly in convex optimization, explains this improved performance by arguing…

Machine Learning · Computer Science 2022-11-07 Kaiqi Jiang , Dhruv Malik , Yuanzhi Li
‹ Prev 1 2 3 10 Next ›