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

arXiv: https://arxiv.org/abs/2505.06699 · HTML: https://mlsanity.com/paper/2505.06699/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 28 of its 38 references that are in the map.
- First posted on arXiv 10 May 2025.

## Builds on (most cited first, 28 of 28)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Distilling the Knowledge in a Neural Network](https://mlsanity.com/paper/1503.02531/) (2015)
- [LAION-400M: Open Dataset of CLIP-Filtered 400 Million Image-Text Pairs](https://mlsanity.com/paper/2111.02114/) (2021)
- [Conceptual 12M: Pushing Web-Scale Image-Text Pre-Training To Recognize Long-Tail Visual Concepts](https://mlsanity.com/paper/2102.08981/) (2021)
- [Reproducible scaling laws for contrastive language-image learning](https://mlsanity.com/paper/2212.07143/) (2022)
- [DataComp: In search of the next generation of multimodal datasets](https://mlsanity.com/paper/2304.14108/) (2023)
- [DeCAF: A Deep Convolutional Activation Feature for Generic Visual Recognition](https://mlsanity.com/paper/1310.1531/) (2013)
- [DoReMi: Optimizing Data Mixtures Speeds Up Language Model Pretraining](https://mlsanity.com/paper/2305.10429/) (2023)
- [Net2Net: Accelerating Learning via Knowledge Transfer](https://mlsanity.com/paper/1511.05641/) (2015)
- [Data Filtering Networks](https://mlsanity.com/paper/2309.17425/) (2023)
- [Prioritized Training on Points that are Learnable, Worth Learning, and Not Yet Learnt](https://mlsanity.com/paper/2206.07137/) (2022)
- [Statistics of Robust Optimization: A Generalized Empirical Likelihood Approach](https://mlsanity.com/paper/1610.03425/) (2016)
- [Large-Scale Methods for Distributionally Robust Optimization](https://mlsanity.com/paper/2010.05893/) (2020)
- [Variance-based regularization with convex objectives](https://mlsanity.com/paper/1610.02581/) (2016)
- [MobileCLIP: Fast Image-Text Models through Multi-Modal Reinforced Training](https://mlsanity.com/paper/2311.17049/) (2023)
- [Perplexed by Perplexity: Perplexity-Based Data Pruning With Small Reference Models](https://mlsanity.com/paper/2405.20541/) (2024)
- [Learning to Grow Pretrained Models for Efficient Transformer Training](https://mlsanity.com/paper/2303.00980/) (2023)
- [Provable Stochastic Optimization for Global Contrastive Learning: Small Batch Does Not Harm Performance](https://mlsanity.com/paper/2202.12387/) (2022)
- [Stacking Your Transformers: A Closer Look at Model Growth for Efficient LLM Pre-Training](https://mlsanity.com/paper/2405.15319/) (2024)
- [Knowledge Inheritance for Pre-trained Language Models](https://mlsanity.com/paper/2105.13880/) (2021)
- [Data curation via joint example selection further accelerates multimodal learning](https://mlsanity.com/paper/2406.17711/) (2024)
- [Bad Students Make Great Teachers: Active Learning Accelerates Large-Scale Visual Understanding](https://mlsanity.com/paper/2312.05328/) (2023)
- [Not All Semantics are Created Equal: Contrastive Self-supervised Learning with Automatic Temperature Individualization](https://mlsanity.com/paper/2305.11965/) (2023)
- [Stochastic Constrained DRO with a Complexity Independent of Sample Size](https://mlsanity.com/paper/2210.05740/) (2022)
- [Non-Smooth Weakly-Convex Finite-sum Coupled Compositional Optimization](https://mlsanity.com/paper/2310.03234/) (2023)
- [Attentional-Biased Stochastic Gradient Descent](https://mlsanity.com/paper/2012.06951/) (2020)
- [FastCLIP: A Suite of Optimization Techniques to Accelerate CLIP Training with Limited Resources](https://mlsanity.com/paper/2407.01445/) (2024)

## Built on by (most cited first, 1 of 1)

- [NeuCLIP: Efficient Large-Scale CLIP Training with Neural Normalizer Optimization](https://mlsanity.com/paper/2511.08417/) (2025)

## Abstract (arXiv)

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. 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, 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.