Summary of Eda-dm: Enhanced Distribution Alignment For Post-training Quantization Of Diffusion Models, by Xuewen Liu et al.
EDA-DM: Enhanced Distribution Alignment for Post-Training Quantization of Diffusion Models
by Xuewen Liu, Zhikai Li, Junrui Xiao, Qingyi Gu
First submitted to arxiv on: 9 Jan 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: 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 proposed Enhanced Distribution Alignment for Post-Training Quantization of Diffusion Models (EDA-DM) aims to address the distribution mismatch issues that plague existing post-training quantization (PTQ) methods for diffusion models. By selecting calibration samples based on density and variety in the latent space, EDA-DM improves alignment at the sample level. Additionally, the modified loss function used during block reconstruction aligns output distributions between the quantized model and its full-precision counterpart at different network granularity levels. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Diffusion models are really good at generating images, but they’re not very fast because they need to do a lot of complicated calculations. To make them faster, scientists tried shrinking the model’s size by using something called post-training quantization (PTQ). However, this didn’t work well for diffusion models because their “thoughts” or activations are super dynamic and hard to compress. The new method, EDA-DM, fixes this problem by choosing the right samples to use during training and adjusting how the model is evaluated during reconstruction. |
Keywords
* Artificial intelligence * Alignment * Diffusion * Latent space * Loss function * Precision * Quantization
