Safe Predictors for Enforcing Input-Output Specifications
arXiv:2001.11062
Abstract
We present an approach for designing correct-by-construction neural networks (and other machine learning models) that are guaranteed to be consistent with a collection of input-output specifications before, during, and after algorithm training. Our method involves designing a constrained predictor for each set of compatible constraints, and combining them safely via a convex combination of their predictions. We demonstrate our approach on synthetic datasets and an aircraft collision avoidance problem.
10 pages, 5 figures, paper accepted to the NeurIPS 2019 Workshop on Machine Learning with Guarantees and the NeurIPS 2019 Workshop on Safety and Robustness in Decision Making
References in corpus (7)
- Adam: A Method for Stochastic Optimization
- Scikit-learn: Machine Learning in Python
- On the Effectiveness of Interval Bound Propagation for Training Verifiably Robust Models
- Evaluating the Robustness of Neural Networks: An Extreme Value Theory Approach
- Towards Proving the Adversarial Robustness of Deep Neural Networks
- Verifying Aircraft Collision Avoidance Neural Networks Through Linear Approximations of Safe Regions
- ART: Abstraction Refinement-Guided Training for Provably Correct Neural Networks