English
Related papers

Related papers: Two-Level K-FAC Preconditioning for Deep Learning

200 papers

Latent variable models are powerful tools for modeling complex phenomena involving in particular partially observed data, unobserved variables or underlying complex unknown structures. Inference is often difficult due to the latent…

Statistics Theory · Mathematics 2023-06-23 Charlotte Baey , Maud Delattre , Estelle Kuhn , Jean-Benoist Leger , Sarah Lemler

Most neural networks are trained using first-order optimization methods, which are sensitive to the parameterization of the model. Natural gradient descent is invariant to smooth reparameterizations because it is defined in a…

Machine Learning · Computer Science 2018-08-31 Kevin Luk , Roger Grosse

Diagonal preconditioners are computationally feasible approximate to second-order optimizers, which have shown significant promise in accelerating training of deep learning models. Two predominant approaches are based on Adam and…

Machine Learning · Computer Science 2025-10-16 Bingbin Liu , Rachit Bansal , Depen Morwani , Nikhil Vyas , David Alvarez-Melis , Sham M. Kakade

Second-order optimization has been developed to accelerate the training of deep neural networks and it is being applied to increasingly larger-scale models. In this study, towards training on further larger scales, we identify a specific…

Machine Learning · Computer Science 2024-06-11 Satoki Ishikawa , Ryo Karakida

Despite the predominant use of first-order methods for training deep learning models, second-order methods, and in particular, natural gradient methods, remain of interest because of their potential for accelerating training through the use…

Machine Learning · Computer Science 2021-12-23 Yi Ren , Donald Goldfarb

Modern deep learning heavily depends on adaptive optimizers such as Adam and its variants, which are renowned for their capacity to handle model scaling and streamline hyperparameter tuning. However, these algorithms typically experience…

Machine Learning · Computer Science 2024-10-18 Son Nguyen , Lizhang Chen , Bo Liu , Qiang Liu

Second-order methods such as KFAC can be useful for neural net training. However, they are often memory-inefficient since their preconditioning Kronecker factors are dense, and numerically unstable in low precision as they require matrix…

Machine Learning · Computer Science 2024-07-24 Wu Lin , Felix Dangel , Runa Eschenhagen , Kirill Neklyudov , Agustinus Kristiadi , Richard E. Turner , Alireza Makhzani

Natural gradients have long been studied in deep reinforcement learning due to their fast convergence properties and covariant weight updates. However, computing natural gradients requires inversion of the Fisher Information Matrix (FIM) at…

Machine Learning · Computer Science 2026-02-12 Yingxiao Huo , Satya Prakash Dash , Radu Stoican , Samuel Kaski , Mingfei Sun

Natural policy gradient methods are popular reinforcement learning methods that improve the stability of policy gradient methods by utilizing second-order approximations to precondition the gradient with the inverse of the…

Machine Learning · Computer Science 2022-10-12 Brennan Gebotys , Alexander Wong , David A. Clausi

A deep neural network is a hierarchical nonlinear model transforming input signals to output signals. Its input-output relation is considered to be stochastic, being described for a given input by a parameterized conditional probability…

Machine Learning · Computer Science 2018-08-23 Shun-ichi Amari , Ryo Karakida , Masafumi Oizumi

First-order methods for stochastic optimization have undeniable relevance, in part due to their pivotal role in machine learning. Variance reduction for these algorithms has become an important research topic. In contrast to common…

Machine Learning · Computer Science 2021-09-08 Manuel Madeira , Renato Negrinho , João Xavier , Pedro M. Q. Aguiar

Multilevel optimization has gained renewed interest in machine learning due to its promise in applications such as hyperparameter tuning and continual learning. However, existing methods struggle with the inherent difficulty of efficiently…

Machine Learning · Computer Science 2024-10-16 Yuntian Gu , Xuzheng Chen

In deep learning, it is common to use more network parameters than training points. In such scenarioof over-parameterization, there are usually multiple networks that achieve zero training error so that thetraining algorithm induces an…

Machine Learning · Computer Science 2023-08-22 Hung-Hsu Chou , Carsten Gieshoff , Johannes Maly , Holger Rauhut

In practical instances of nonconvex matrix factorization, the rank of the true solution $r^{\star}$ is often unknown, so the rank $r$ of the model can be overspecified as $r>r^{\star}$. This over-parameterized regime of matrix factorization…

Optimization and Control · Mathematics 2025-04-15 Gavin Zhang , Salar Fattahi , Richard Y. Zhang

Pipeline parallelism enables efficient training of Large Language Models (LLMs) on large-scale distributed accelerator clusters. Yet, pipeline bubbles during startup and tear-down reduce the utilization of accelerators. Although efficient…

Machine Learning · Computer Science 2023-05-16 Kazuki Osawa , Shigang Li , Torsten Hoefler

This paper introduces a new stochastic optimization method based on the regularized Fisher information matrix (FIM), named SOFIM, which can efficiently utilize the FIM to approximate the Hessian matrix for finding Newton's gradient update…

Machine Learning · Computer Science 2024-05-02 Mrinmay Sen , A. K. Qin , Gayathri C , Raghu Kishore N , Yen-Wei Chen , Balasubramanian Raman

We propose a new algorithm for efficiently solving the damped Fisher matrix in large-scale scenarios where the number of parameters significantly exceeds the number of available samples. This problem is fundamental for natural gradient…

Machine Learning · Computer Science 2023-10-27 Yixiao Chen , Hao Xie , Han Wang

Convolutional Networks (ConvNets) have recently improved image recognition performance thanks to end-to-end learning of deep feed-forward models from raw pixels. Deep learning is a marked departure from the previous state of the art, the…

Computer Vision and Pattern Recognition · Computer Science 2015-07-24 Albert Gordo , Adrien Gaidon , Florent Perronnin

This work proposes a time-efficient Natural Gradient Descent method, called TENGraD, with linear convergence guarantees. Computing the inverse of the neural network's Fisher information matrix is expensive in NGD because the Fisher matrix…

Machine Learning · Computer Science 2022-03-04 Saeed Soori , Bugra Can , Baourun Mu , Mert Gürbüzbalaban , Maryam Mehri Dehnavi

Adaptive gradient methods like Adagrad and its variants are widespread in large-scale optimization. However, their use of diagonal preconditioning matrices limits the ability to capture parameter correlations. Full-matrix adaptive methods,…

Machine Learning · Computer Science 2025-09-01 Tatyana Matveeva , Aleksandr Katrutsa , Evgeny Frolov