Beyond Predictions in Neural ODEs: Identification and Interventions
Abstract: Spurred by tremendous success in pattern matching and prediction tasks, researchers increasingly resort to machine learning to aid original scientific discovery. Given large amounts of observational data about a system, can we uncover the rules that govern its evolution? Solving this task holds the great promise of fully understanding the causal interactions and being able to make reliable predictions about the system’s behavior under interventions. We take a step towards such system identification for time-series data generated from systems
of ordinary differential equations (ODEs) using flexible neural ODEs. Neural ODEs have proven successful in learning dynamical systems in terms of recovering observed trajectories. However, their efficacy in learning ground truth dynamics and making predictions under
unseen interventions are still underexplored. We develop a simple regularization scheme for neural ODEs that helps in recovering the dynamics and causal structure from time-series data. Our results on a variety of (non)-linear first and second order systems as well as
real data validate our method. We conclude by showing that we can also make accurate predictions under interventions on variables or the system itself.
Submission Length: Regular submission (no more than 12 pages of main content)
Assigned Action Editor: ~Mauricio_A_Álvarez1
Submission Number: 447
Loading