Saved in:
Bibliographic Details
Main Authors: Rukhovich, Alexey, Podolskiy, Alexander, Piontkovskaya, Irina
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2501.15556
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910801413013504
author Rukhovich, Alexey
Podolskiy, Alexander
Piontkovskaya, Irina
author_facet Rukhovich, Alexey
Podolskiy, Alexander
Piontkovskaya, Irina
contents In multi-domain learning, a single model is trained on diverse data domains to leverage shared knowledge and improve generalization. The order in which the data from these domains is used for training can significantly affect the model's performance on each domain. However, this dependence is under-studied. In this paper, we investigate the influence of training order (or data mixing) in multi-domain learning using the concept of Lie bracket of gradient vector fields. By analyzing the infinitesimal effects of changing the training order, we identify regions in the parameter space where altering the order between two training domains can benefit the target loss. We validate the predictions of our theoretical framework on the influence of training order (or data mixing) both on a toy example and bilingual LLM pre-training.
format Preprint
id arxiv_https___arxiv_org_abs_2501_15556
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Commute Your Domains: Trajectory Optimality Criterion for Multi-Domain Learning
Rukhovich, Alexey
Podolskiy, Alexander
Piontkovskaya, Irina
Machine Learning
Computation and Language
In multi-domain learning, a single model is trained on diverse data domains to leverage shared knowledge and improve generalization. The order in which the data from these domains is used for training can significantly affect the model's performance on each domain. However, this dependence is under-studied. In this paper, we investigate the influence of training order (or data mixing) in multi-domain learning using the concept of Lie bracket of gradient vector fields. By analyzing the infinitesimal effects of changing the training order, we identify regions in the parameter space where altering the order between two training domains can benefit the target loss. We validate the predictions of our theoretical framework on the influence of training order (or data mixing) both on a toy example and bilingual LLM pre-training.
title Commute Your Domains: Trajectory Optimality Criterion for Multi-Domain Learning
topic Machine Learning
Computation and Language
url https://arxiv.org/abs/2501.15556