Summary of Self-consuming Generative Models with Curated Data Provably Optimize Human Preferences, by Damien Ferbach et al.
Self-Consuming Generative Models with Curated Data Provably Optimize Human Preferences
by Damien Ferbach, Quentin Bertrand, Avishek Joey Bose, Gauthier Gidel
First submitted to arxiv on: 12 Jun 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); 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 The paper investigates the impact of data curation on iterated retraining of generative models, particularly in scenarios where users provide human feedback. It proposes an implicit preference optimization mechanism, which allows the model to learn from curated data without access to reward functions or negative samples. The authors demonstrate that this approach maximizes the expected reward and provides stability when using a positive fraction of real data at each retraining step. They also conduct experiments on synthetic datasets and CIFAR10, showing that the procedure amplifies biases in the reward model. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper looks at how people curate data for generative models, like Stable Diffusion or Midjourney, which produces multiple image variations based on user queries. It explores whether this curation affects how well the model works when it’s retrained using some of the curated data. The researchers show that this process can be seen as an optimization mechanism, even though the model doesn’t know the rules for what makes good or bad data. They also find that when you use a mix of real and curated data, the process becomes more stable. |
Keywords
» Artificial intelligence » Diffusion » Optimization
