Summary of Group Distributionally Robust Dataset Distillation with Risk Minimization, by Saeed Vahidian et al.
Group Distributionally Robust Dataset Distillation with Risk Minimization
by Saeed Vahidian, Mingyu Wang, Jianyang Gu, Vyacheslav Kungurtsev, Wei Jiang, Yiran Chen
First submitted to arxiv on: 7 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
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 proposed algorithm for dataset distillation (DD) aims to create a synthetic dataset that captures the essential information of a training dataset, allowing for the training of accurate neural models. The traditional methods for constructing synthetic data rely on matching the convergence properties of training with the synthetic and training datasets. However, these methods overlook the relationship between DD and its generalization, particularly across uncommon subgroups. To address this issue, the authors introduce an algorithm that combines clustering with the minimization of a risk measure on the loss to conduct DD. Theoretical justifications are provided, and numerical experiments demonstrate the effective generalization and robustness of the approach across subgroups. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Dataset distillation creates a synthetic dataset that helps train accurate neural models. It’s like making a copy of your original data, but instead of copying everything, you keep what matters most. This can help with things like transfer learning and training models on hard-to-reach areas. But so far, nobody has looked at how well this copied data does when it meets new, different samples. That’s what this paper is all about – making a better copied dataset that works well even in tricky situations. |
Keywords
* Artificial intelligence * Clustering * Distillation * Generalization * Synthetic data * Transfer learning
