English

Model Steering: Learning with a Reference Model Improves Generalization Bounds and Scaling Laws

Machine Learning 2025-05-21 v3 Artificial Intelligence Computer Vision and Pattern Recognition Machine Learning

Abstract

This paper formalizes an emerging learning paradigm that uses a trained model as a reference to guide and enhance the training of a target model through strategic data selection or weighting, named model steering\textbf{model steering}. While ad-hoc methods have been used in various contexts, including the training of large foundation models, its underlying principles remain insufficiently understood, leading to sub-optimal performance. In this work, we propose a theory-driven framework for model steering called DRRho risk minimization\textbf{DRRho risk minimization}, which is rooted in Distributionally Robust Optimization (DRO). Through a generalization analysis, we provide theoretical insights into why this approach improves generalization and data efficiency compared to training without a reference model. To the best of our knowledge, this is the first time such theoretical insights are provided for the new learning paradigm, which significantly enhance our understanding and practice of model steering. Building on these insights and the connection between contrastive learning and DRO, we introduce a novel method for Contrastive Language-Image Pretraining (CLIP) with a reference model, termed DRRho-CLIP. Extensive experiments validate the theoretical insights, reveal a superior scaling law compared to CLIP without a reference model, and demonstrate its strength over existing heuristic approaches.

Keywords

Cite

@article{arxiv.2505.06699,
  title  = {Model Steering: Learning with a Reference Model Improves Generalization Bounds and Scaling Laws},
  author = {Xiyuan Wei and Ming Lin and Fanjiang Ye and Fengguang Song and Liangliang Cao and My T. Thai and Tianbao Yang},
  journal= {arXiv preprint arXiv:2505.06699},
  year   = {2025}
}

Comments

18 pages, 6 figures

R2 v1 2026-06-28T23:28:14.205Z