Summary of Unsupervised Discovery Of Steerable Factors When Graph Deep Generative Models Are Entangled, by Shengchao Liu et al.
Unsupervised Discovery of Steerable Factors When Graph Deep Generative Models Are Entangled
by Shengchao Liu, Chengpeng Wang, Jiarui Lu, Weili Nie, Hanchen Wang, Zhuoxinran Li, Bolei Zhou, Jian Tang
First submitted to arxiv on: 29 Jan 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); 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 proposed method, GraphCG, aims to uncover the latent space of pre-trained deep generative models (DGMs) for graph data. By examining the representation space of three pretrained graph DGMs using six disentanglement metrics, researchers found that the space was entangled. To address this issue, GraphCG learns steerable factors by maximizing mutual information between semantic-rich directions, allowing controlled graph movement along the same direction to share the same factors. Experimental results demonstrate GraphCG’s superiority over four competitive baselines on two graph DGMs pretrained on molecule datasets. Furthermore, qualitative analysis illustrates seven learned steerable factors across five pre-trained DGMs and five graph datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Graph CG is a new way to understand the hidden space inside big computers that can generate complex pictures of molecules. Right now, these computers are very good at making fake molecules, but they’re not very good at controlling what kind of molecule they make. Graph CG helps by finding special directions in the computer’s mind that it uses to create different kinds of molecules. By studying how these directions work, scientists can make better computers that can generate new and interesting molecules. |
Keywords
* Artificial intelligence * Latent space
