Summary of Distribution Learning For Molecular Regression, by Nima Shoghi et al.
Distribution Learning for Molecular Regression
by Nima Shoghi, Pooya Shoghi, Anuroop Sriram, Abhishek Das
First submitted to arxiv on: 30 Jul 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Quantitative Methods (q-bio.QM)
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 paper explores the application of “soft” targets in machine learning for regression tasks, specifically in molecular property prediction. Current methods using soft targets for regression are limited by biases and lack empirical evaluation. This work assesses existing methods’ strengths and drawbacks on molecular property regression tasks, identifying key biases and proposing methods to address them through ablation studies. The authors introduce Distributional Mixture of Experts (DMoE), a model-independent method that trains a model to predict probability distributions of targets. DMoE’s loss function combines cross-entropy and L1 distance to produce a robust solution. The paper evaluates DMoE on multiple datasets (Open Catalyst, MD17, QM9) and architectures (SchNet, GemNet, Graphormer), showing improvements over baselines across all combinations. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how to make machines better at predicting things about molecules. Right now, people are using a technique called “soft targets” to improve classification (yes or no questions) but haven’t applied it much to regression (estimating values). The authors investigate what works and what doesn’t with this method for predicting molecular properties. They propose a new approach called DMoE that’s better at handling some of the limitations of current methods. They test DMoE on different datasets and models, showing that it can do a better job than existing methods. |
Keywords
» Artificial intelligence » Classification » Cross entropy » Loss function » Machine learning » Mixture of experts » Probability » Regression
