arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2406.07515cs.LGcs.AIstat.ML
收藏

超越模型崩溃:利用合成数据进行扩展需要验证

Beyond Model Collapse: Scaling Up with Synthesized Data Requires Verification

Meta FAIR;Center for Data Science, New York University(纽约大学数据科学中心);Courant Institue of Mathematical Sciences, New York University(纽约大学柯朗数学科学研究所);School of Mathematical Sciences, Peking University(北京大学数学科学学院)

Yunzhen Feng, Elvis Dohmatob, Pu Yang, Francois Charton, Julia Kempe

AI总结:

针对合成数据训练导致的模型崩溃问题,论文提出用验证器筛选数据,并通过高斯混合理论与两项任务实验证明不完美验证器也能有效防止性能下降。

摘要:

大型语言模型(LLM)越来越多地使用由其他 LLM 生成的数据进行训练,这既因为生成的文本和图像会成为预训练语料库的一部分,也因为合成数据被用来替代昂贵的人工标注。这引发了人们对“模型崩溃”的担忧,即当模型的训练集包含生成数据时,其性能会下降。考虑到人类和机器区分好样本与坏样本都比生成高质量样本更容易,我们研究了对合成数据使用验证以防止模型崩溃。我们利用高斯混合模型、线性分类器和线性验证器给出了理论刻画,推导出带有可测量代理指标的条件,以评估验证器能否有效筛选出可带来最优性能的合成数据。我们在两个实际任务上进行了实验——使用 transformer 计算矩阵特征值,以及使用 LLM 进行新闻摘要——二者在使用生成数据训练时都会出现模型崩溃;结果表明,即使是不完美的验证器,也确实可以被用来防止模型崩溃,并且我们提出的代理度量与性能具有很强的相关性。

英文摘要:

Large Language Models (LLM) are increasingly trained on data generated by other LLM, either because generated text and images become part of the pre-training corpus, or because synthetized data is used as a replacement for expensive human-annotation. This raises concerns about \emph{model collapse}, a drop in model performance when their training sets include generated data. Considering that it is easier for both humans and machines to tell between good and bad examples than to generate high-quality samples, we investigate the use of verification on synthesized data to prevent model collapse. We provide a theoretical characterization using Gaussian mixtures, linear classifiers, and linear verifiers to derive conditions with measurable proxies to assess whether the verifier can effectively select synthesized data that leads to optimal performance. We experiment with two practical tasks -- computing matrix eigenvalues with transformers and news summarization with LLMs -- which both exhibit model collapse when trained on generated data, and show that verifiers, even imperfect ones, can indeed be harnessed to prevent model collapse and that our proposed proxy measure strongly correlates with performance.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑