paper-with-me

홈 › Papers

Efficient Hyperparameter Tuning via Trajectory Invariance Principle

2025-09-29 · Bingrui Li, Jiaxin Wen, Zhanpeng Zhou, Jun Zhu, Jianfei Chen arxiv

As hyperparameter tuning becomes increasingly costly at scale, efficient tuning methods are essential. Yet principles for guiding hyperparameter tuning remain limited. In this work, we seek to establish such principles by considering a broad range of hyperparameters, including batch size, learning rate, and weight decay. We identify a phenomenon we call trajectory invariance, where pre-training loss curves, gradient noise, and gradient norm exhibit invariance--closely overlapping--with respect to a quantity that combines learning rate and weight decay. This phenomenon effectively reduces the original two-dimensional hyperparameter space to one dimension, yielding an efficient tuning rule: follow the salient direction revealed by trajectory invariance. Furthermore, we refine previous scaling laws and challenge several existing viewpoints. Overall, our work proposes new principles for efficient tuning and inspires future research on scaling laws.

📄 PDF Abstract BibTeX arXiv:2509.25049

Code (0)

등록된 구현이 없습니다.

Similar Papers 제목 키워드 기반

Principled Design of Diffusion-based Optimizers for Inverse Problems

2026-05-12 · Julio Oscanoa, Irmak Sivgin, Cagan Alkan, Daniel Ennis 외 arxiv

Score-based diffusion models achieve state-of-the-art performance for inverse problems, but their practical deployment is hindered by long inference times and cumbersome hyperparameter tuning. While pretrained diffusion …

Image Reconstruction

Optimizing Training Trajectories in Variational Autoencoders via Latent Bayesian Optimization Approach

2022-06-30 · Arpan Biswas, Rama Vasudevan, Maxim Ziatdinov, Sergei V. Kalinin

Unsupervised and semi-supervised ML methods such as variational autoencoders (VAE) have become widely adopted across multiple areas of physics, chemistry, and materials sciences due to their capability in disentangling r…

Bayesian Optimization

Towards Hyperparameter-Agnostic DNN Training via Dynamical System Insights

2023-10-21 · Carmel Fiscko, Aayushya Agarwal, Yihan Ruan, Soummya Kar 외

We present a stochastic first-order optimization method specialized for deep neural networks (DNNs), ECCO-DNN. This method models the optimization variable trajectory as a dynamical system and develops a discretization a…

Numerical Integration

Parameter Tuning with Generalization Guarantees for GPU-Accelerated Linear Programming

2026-06-07 · Siddharth Prasad, Dravyansh Sharma arxiv

Recent research has developed practical, parallelizable first-order methods for large scale linear programming, but performance is highly dependent on hyperparameter selection. We derive generalization guarantees for hyp…

CMA-ES for Hyperparameter Optimization of Deep Neural Networks

2016-04-25 · Ilya Loshchilov, Frank Hutter

Hyperparameters of deep neural networks are often optimized by grid search, random search or Bayesian optimization. As an alternative, we propose to use the Covariance Matrix Adaptation Evolution Strategy (CMA-ES), which…

Bayesian OptimizationHyperparameter Optimization