Summary of Self-correcting Self-consuming Loops For Generative Model Training, by Nate Gillman et al.
Self-Correcting Self-Consuming Loops for Generative Model Training
by Nate Gillman, Michael Freeman, Daksh Aggarwal, Chia-Hong Hsu, Calvin Luo, Yonglong Tian, Chen Sun
First submitted to arxiv on: 11 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); 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 Machine learning models are increasingly trained on a mix of human- and machine-generated data. While successful for representation learning, using synthetic data for generative model training can create “self-consuming loops” that may lead to training instability or collapse unless certain conditions are met. Our paper aims to stabilize self-consuming generative model training by introducing an idealized correction function that maps a data point to be more likely under the true data distribution. We propose self-correction functions that rely on expert knowledge and aim to approximate the idealized corrector automatically and at scale. Empirical validation on the challenging human motion synthesis task demonstrates the effectiveness of self-correcting self-consuming loops in avoiding model collapse, even when synthetic data makes up 100% of the training set. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Machine learning models are trained with a mix of real and fake data. Fake data can help, but it also creates problems that need to be fixed. Our research shows how to make this process more stable by using special functions that correct for mistakes in the fake data. We tested our method on making realistic human movements look like they were done by real people, and it worked even when most of the training data was fake. |
Keywords
* Artificial intelligence * Generative model * Machine learning * Representation learning * Synthetic data
