Statistics > Machine Learning
[Submitted on 13 Feb 2018 (v1), last revised 6 Jun 2018 (this version, v2)]
Title:Neural Relational Inference for Interacting Systems
Download PDFAbstract:Interacting systems are prevalent in nature, from dynamical systems in physics to complex societal dynamics. The interplay of components can give rise to complex behavior, which can often be explained using a simple model of the system's constituent parts. In this work, we introduce the neural relational inference (NRI) model: an unsupervised model that learns to infer interactions while simultaneously learning the dynamics purely from observational data. Our model takes the form of a variational auto-encoder, in which the latent code represents the underlying interaction graph and the reconstruction is based on graph neural networks. In experiments on simulated physical systems, we show that our NRI model can accurately recover ground-truth interactions in an unsupervised manner. We further demonstrate that we can find an interpretable structure and predict complex dynamics in real motion capture and sports tracking data.
Submission history
From: Thomas Kipf [view email][v1] Tue, 13 Feb 2018 15:35:11 UTC (2,401 KB)
[v2] Wed, 6 Jun 2018 14:02:33 UTC (3,115 KB)
Current browse context:
stat.ML
References & Citations
Bibliographic and Citation Tools
Bibliographic Explorer (What is the Explorer?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)