Summary of Label-noise Robust Diffusion Models, by Byeonghu Na et al.
Label-Noise Robust Diffusion Models
by Byeonghu Na, Yeongmin Kim, HeeSun Bae, Jung Hyun Lee, Se Jung Kwon, Wanmo Kang, Il-Chul Moon
First submitted to arxiv on: 27 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 This paper proposes a novel approach to training conditional diffusion models when dealing with noisy labels, a common issue in large-scale datasets. The proposed method, Transition-aware weighted Denoising Score Matching (TDSM), is designed to address the condition mismatch and quality degradation caused by noise in conditional inputs. TDSM incorporates instance-wise and time-dependent label transition probabilities into its objective function, which includes a weighted sum of score networks. The method also introduces a transition-aware weight estimator that leverages a time-dependent noisy-label classifier customized for the diffusion process. Experiments across various datasets and noisy label settings demonstrate the effectiveness of TDSM in improving the quality of generated samples aligned with given conditions. Additionally, TDSM shows improved generation performance on prevalent benchmark datasets, highlighting its potential as a part of label-noise robust generative models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research paper helps make computers that can create new data based on some rules. These computers need to learn from big datasets, but sometimes these datasets have mistakes in the rules they’re supposed to follow. This makes it harder for the computer to create good data. The researchers found a way to fix this problem by creating a new method called TDSM. It’s like a special filter that helps the computer ignore the mistakes and focus on the correct information. They tested their method with different datasets and showed that it works better than other approaches. This is important because it can help computers create more realistic data, which has many applications in fields like medicine, science, and art. |
Keywords
* Artificial intelligence * Diffusion * Objective function
