Summary of Rate Of Model Collapse in Recursive Training, by Ananda Theertha Suresh and Andrew Thangaraj and Aditya Nanda Kishore Khandavally
Rate of Model Collapse in Recursive Training
by Ananda Theertha Suresh, Andrew Thangaraj, Aditya Nanda Kishore Khandavally
First submitted to arxiv on: 23 Dec 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Information Theory (cs.IT); Machine Learning (stat.ML)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary In this research paper, the authors investigate how machine learning models’ ability to capture nuances in original human-generated data degrades over time when recursively trained on generated data from previous rounds. They focus on well-studied distribution families like discrete and Gaussian distributions, exploring the rate of model collapse under maximum likelihood estimation during recursive training. The study theoretically characterizes the rate of collapse for these fundamental settings and validates findings with experimental evaluations. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Machine learning models can be trained on synthetic data created by previous models, leading to concerns about their quality. The authors look at how fast this degradation occurs when training models recursively on generated data. They find that even for simple distributions like discrete and Gaussian, the exact rate of model collapse is unknown. By studying these fundamental settings under maximum likelihood estimation, they show that the time it takes for a model to forget information depends on the number of times that information appeared in the original data. |
Keywords
» Artificial intelligence » Likelihood » Machine learning » Synthetic data
