Summary of Diffinject: Revisiting Debias Via Synthetic Data Generation Using Diffusion-based Style Injection, by Donggeun Ko et al.
DiffInject: Revisiting Debias via Synthetic Data Generation using Diffusion-based Style Injection
by Donggeun Ko, Sangwoo Jo, Dongjun Lee, Namjun Park, Jaekwang Kim
First submitted to arxiv on: 10 Jun 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); 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 A novel machine learning approach, called DiffInject, is proposed to address dataset bias in image-based models. This bias occurs when specific attributes like texture or color are learned unintentionally, resulting in poor performance. Previous efforts have focused on debiasing algorithms or generating synthetic data to mitigate biases. However, current generative approaches rely heavily on scarce bias-specific samples from the dataset. DiffInject uses a pretrained diffusion model to augment synthetic bias-conflict samples in the latent space, without requiring explicit knowledge of bias types or labelling. This fully unsupervised framework demonstrates significant results in reducing dataset bias. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary A new machine learning method helps reduce biases in image-based models by creating fake data that makes these models fairer. Biases happen when models learn specific details like texture or color, which can make them perform poorly. To fix this, researchers have tried using special algorithms or making synthetic data to cancel out biases. But current methods rely on very little information about the bias and don’t work well. The new approach, called DiffInject, uses a special kind of AI model to create fake data that helps models be more fair. |
Keywords
» Artificial intelligence » Diffusion model » Latent space » Machine learning » Synthetic data » Unsupervised
