GoTube: Scalable Statistical Verification of Continuous-Depth Models

Authors

  • Sophie A. Gruenbacher TU Wien
  • Mathias Lechner IST Austria
  • Ramin Hasani MIT
  • Daniela Rus MIT CSAIL
  • Thomas A. Henzinger IST Austria
  • Scott A. Smolka Stony Brook University
  • Radu Grosu TU Wien

DOI:

https://doi.org/10.1609/aaai.v36i6.20631

Keywords:

Machine Learning (ML)

Abstract

We introduce a new statistical verification algorithm that formally quantifies the behavioral robustness of any time-continuous process formulated as a continuous-depth model. Our algorithm solves a set of global optimization (Go) problems over a given time horizon to construct a tight enclosure (Tube) of the set of all process executions starting from a ball of initial states. We call our algorithm GoTube. Through its construction, GoTube ensures that the bounding tube is conservative up to a desired probability and up to a desired tightness. GoTube is implemented in JAX and optimized to scale to complex continuous-depth neural network models. Compared to advanced reachability analysis tools for time-continuous neural networks, GoTube does not accumulate overapproximation errors between time steps and avoids the infamous wrapping effect inherent in symbolic techniques. We show that GoTube substantially outperforms state-of-the-art verification tools in terms of the size of the initial ball, speed, time-horizon, task completion, and scalability on a large set of experiments. GoTube is stable and sets the state-of-the-art in terms of its ability to scale to time horizons well beyond what has been previously possible.

Downloads

Published

2022-06-28

How to Cite

Gruenbacher, S. A., Lechner, M., Hasani, R., Rus, D., Henzinger, T. A., Smolka, S. A., & Grosu, R. (2022). GoTube: Scalable Statistical Verification of Continuous-Depth Models. Proceedings of the AAAI Conference on Artificial Intelligence, 36(6), 6755-6764. https://doi.org/10.1609/aaai.v36i6.20631

Issue

Section

AAAI Technical Track on Machine Learning I