Nathan’s paper on “Almost Surely Stable Deep Dynamics” has been accepted at NeurIPS 2020 and selected as a Spotlight session. Congratulations! At NeurIPS 2020, 385 papers out of 1900 were selected as spotlights or orals. 1900 papers were accepted out of ~11,000 submissions.
Almost Surely Stable Deep Dynamics by Nathan P. Lawrence, Philip D. Loewen, Michael G. Forbes, Johan U. Backstrom and R. Bhushan Gopaluni
Abstract
We introduce a method for learning provably stable deep neural network based dynamic models from observed data. Specifically, we consider discrete-time stochastic dynamic models, as they are of particular interest in practical applications such as estimation and control. However, these aspects exacerbate the challenge of guaranteeing stability of a neural network dynamic model. Our method constrains the dynamic model to be stable subject to a neural network Lyapunov function. To this end, we propose two approaches: one exploits convexity of the Lyapunov function, while the other enforces stability through an implicit output layer. Numerical results are presented and the accompanying code is (will be) publicly available.
-
Read the pre-print: 2020C6_Lawrence_NeurIPS.pdf
-
Poster available on Figshare: Almost Surely Stable Deep Dynamics Poster