Saved in:
Bibliographic Details
Main Authors: Yang, Pu, Feng, Yunzhen, Chen, Ziyuan, Wu, Yuhang, Li, Zhuoyuan
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2501.18962
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911212612091904
author Yang, Pu
Feng, Yunzhen
Chen, Ziyuan
Wu, Yuhang
Li, Zhuoyuan
author_facet Yang, Pu
Feng, Yunzhen
Chen, Ziyuan
Wu, Yuhang
Li, Zhuoyuan
contents Modern foundation models often undergo iterative ``bootstrapping'' in their post-training phase: a model generates synthetic data, an external verifier filters out low-quality samples, and the high-quality subset is used for further fine-tuning. Over multiple iterations, the model performance improves, raising a crucial question: How should the total budget for generation and training be allocated across iterations to maximize final performance? In this work, we develop a theoretical framework for analyzing budget allocation strategies. Specifically, we show that constant policies fail to converge with high probability, while increasing policies -- particularly exponential growth policies -- exhibit significant theoretical advantages. Experiments on image denoising with diffusion probabilistic models and math reasoning with large language models show that both exponential and polynomial growth policies consistently outperform constant policies, with exponential policies often providing more stable performance.
format Preprint
id arxiv_https___arxiv_org_abs_2501_18962
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Spend Wisely: Maximizing Post-Training Gains in Iterative Synthetic Data Bootstrapping
Yang, Pu
Feng, Yunzhen
Chen, Ziyuan
Wu, Yuhang
Li, Zhuoyuan
Machine Learning
Modern foundation models often undergo iterative ``bootstrapping'' in their post-training phase: a model generates synthetic data, an external verifier filters out low-quality samples, and the high-quality subset is used for further fine-tuning. Over multiple iterations, the model performance improves, raising a crucial question: How should the total budget for generation and training be allocated across iterations to maximize final performance? In this work, we develop a theoretical framework for analyzing budget allocation strategies. Specifically, we show that constant policies fail to converge with high probability, while increasing policies -- particularly exponential growth policies -- exhibit significant theoretical advantages. Experiments on image denoising with diffusion probabilistic models and math reasoning with large language models show that both exponential and polynomial growth policies consistently outperform constant policies, with exponential policies often providing more stable performance.
title Spend Wisely: Maximizing Post-Training Gains in Iterative Synthetic Data Bootstrapping
topic Machine Learning
url https://arxiv.org/abs/2501.18962