Summary of Controlled Training Data Generation with Diffusion Models, by Teresa Yeo et al.
Controlled Training Data Generation with Diffusion Models
by Teresa Yeo, Andrei Atanov, Harold Benoit, Aleksandr Alekseev, Ruchira Ray, Pooya Esmaeil Akhoondi, Amir Zamir
First submitted to arxiv on: 22 Mar 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Computation and Language (cs.CL); Machine Learning (cs.LG)
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 researchers present a method to control a text-to-image generative model for producing training data that is useful for supervised learning. Unlike previous works, this approach uses an automated closed-loop system with two feedback mechanisms to generate diverse data informed by a given supervised model. The first mechanism finds adversarial prompts that maximize the model loss, while the second mechanism guides the generation process towards a target distribution. The proposed method, Guided Adversarial Prompts, is evaluated on various tasks, datasets, and architectures with different types of distribution shifts, demonstrating its efficiency compared to open-loop approaches. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps us create more useful training data for machines that learn from images. Instead of using random prompts, they developed a system that learns what to say to get the best results. They showed how this method works well on different tasks and datasets, even when the pictures look very different from what the model has seen before. |
Keywords
* Artificial intelligence * Generative model * Supervised
