Summary of One Category One Prompt: Dataset Distillation Using Diffusion Models, by Ali Abbasi et al.
One Category One Prompt: Dataset Distillation using Diffusion Models
by Ali Abbasi, Ashkan Shahbazi, Hamed Pirsiavash, Soheil Kolouri
First submitted to arxiv on: 11 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 paper introduces Dataset Distillation using Diffusion Models (D3M), a novel approach to dataset distillation that leverages recent advancements in generative text-to-image foundation models. The traditional dataset distillation methods struggle to scale effectively due to bi-level optimization limitations, whereas D3M exploits knowledge distillation with decoupled optimization schemes to address scalability issues. Our method uses textual inversion for fine-tuning text-to-image generative models, creating concise and informative representations for large datasets. This allows efficient storage and inference of new samples within a fixed memory budget, demonstrated through extensive experiments across various computer vision benchmark datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about finding a way to make big datasets smaller without losing important information. Right now, it’s hard to store and move these big datasets because they take up too much space. The researchers came up with an idea called D3M (Dataset Distillation using Diffusion Models). It uses special computer models that can generate pictures from text descriptions. This helps create a small version of the dataset that still has all the important details. They tested this method on some famous image recognition tests and it worked really well, even when they had limited space. |
Keywords
* Artificial intelligence * Distillation * Fine tuning * Inference * Knowledge distillation * Optimization
