Summary of Mitigating Exposure Bias in Score-based Generation Of Molecular Conformations, by Sijia Wang et al.
Mitigating Exposure Bias in Score-Based Generation of Molecular Conformations
by Sijia Wang, Chen Wang, Zhenhao Zhao, Jiqiang Zhang, Weiran Cai
First submitted to arxiv on: 21 Sep 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Biomolecules (q-bio.BM)
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 approach to molecular conformation generation using Score-Based Generative Models (SGMs) is proposed, addressing the long-standing issue of exposure bias in these models. SGMs are effective for generating accurate conformations, but the discrepancy between training and inference leads to exposure bias. A method is developed to measure this bias in SGMs, confirming its significant existence and providing a value. To compensate for exposure bias, a new algorithm called Input Perturbation (IP) is introduced, adapted from a method originally designed for Diffusion Probabilistic Models (DPMs). Experimental results demonstrate that IP-enhanced SGM-based models can significantly improve accuracy and diversity of generated conformations. Notably, the IP-enhanced Torsional Diffusion model achieves state-of-the-art performance on the GEOM-Drugs dataset and is on par with existing methods on GEOM-QM9. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Molecular conformation generation is a big challenge in chemistry. Recently, new models called SGMs have been really good at generating accurate shapes of molecules. But there’s a problem – these models don’t always work as well when they’re actually used to make predictions. This is because the way they learn and the way they make predictions are different. The paper proposes a solution to this problem by measuring how much this difference affects the results. They also develop a new technique called Input Perturbation that helps fix this issue. By using this technique, the models can generate even more accurate and diverse shapes of molecules. |
Keywords
* Artificial intelligence * Diffusion * Diffusion model * Inference
