Infinite time horizon safety of Bayesian neural networks
Lechner M, Žikelić Ð, Chatterjee K, Henzinger TA. 2021. Infinite time horizon safety of Bayesian neural networks. 35th Conference on Neural Information Processing Systems. NeurIPS: Neural Information Processing Systems, Advances in Neural Information Processing Systems, .
Download
Download (ext.)
Conference Paper
| Published
| English
Author
Grant
Series Title
Advances in Neural Information Processing Systems
Abstract
Bayesian neural networks (BNNs) place distributions over the weights of a neural network to model uncertainty in the data and the network's prediction. We consider the problem of verifying safety when running a Bayesian neural network policy in a feedback loop with infinite time horizon systems. Compared to the existing sampling-based approaches, which are inapplicable to the infinite time horizon setting, we train a separate deterministic neural network that serves as an infinite time horizon safety certificate. In particular, we show that the certificate network guarantees the safety of the system over a subset of the BNN weight posterior's support. Our method first computes a safe weight set and then alters the BNN's weight posterior to reject samples outside this set. Moreover, we show how to extend our approach to a safe-exploration reinforcement learning setting, in order to avoid unsafe trajectories during the training of the policy. We evaluate our approach on a series of reinforcement learning benchmarks, including non-Lyapunovian safety specifications.
Publishing Year
Date Published
2021-12-01
Proceedings Title
35th Conference on Neural Information Processing Systems
Acknowledgement
This research was supported in part by the Austrian Science Fund (FWF) under grant Z211-N23 (Wittgenstein Award), ERC CoG 863818 (FoRM-SMArt), and the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie Grant Agreement No. 665385.
Conference
NeurIPS: Neural Information Processing Systems
Conference Location
Virtual
Conference Date
2021-12-06 – 2021-12-10
IST-REx-ID
Cite this
Lechner M, Žikelić Ð, Chatterjee K, Henzinger TA. Infinite time horizon safety of Bayesian neural networks. In: 35th Conference on Neural Information Processing Systems. ; 2021. doi:10.48550/arXiv.2111.03165
Lechner, M., Žikelić, Ð., Chatterjee, K., & Henzinger, T. A. (2021). Infinite time horizon safety of Bayesian neural networks. In 35th Conference on Neural Information Processing Systems. Virtual. https://doi.org/10.48550/arXiv.2111.03165
Lechner, Mathias, Ðorđe Žikelić, Krishnendu Chatterjee, and Thomas A Henzinger. “Infinite Time Horizon Safety of Bayesian Neural Networks.” In 35th Conference on Neural Information Processing Systems, 2021. https://doi.org/10.48550/arXiv.2111.03165.
M. Lechner, Ð. Žikelić, K. Chatterjee, and T. A. Henzinger, “Infinite time horizon safety of Bayesian neural networks,” in 35th Conference on Neural Information Processing Systems, Virtual, 2021.
Lechner M, Žikelić Ð, Chatterjee K, Henzinger TA. 2021. Infinite time horizon safety of Bayesian neural networks. 35th Conference on Neural Information Processing Systems. NeurIPS: Neural Information Processing Systems, Advances in Neural Information Processing Systems, .
Lechner, Mathias, et al. “Infinite Time Horizon Safety of Bayesian Neural Networks.” 35th Conference on Neural Information Processing Systems, 2021, doi:10.48550/arXiv.2111.03165.
All files available under the following license(s):
Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported (CC BY-NC-ND 3.0):
Main File(s)
File Name
infinite_time_horizon_safety_o.pdf
452.49 KB
Access Level
Open Access
Date Uploaded
2022-01-26
MD5 Checksum
0fc0f852525c10dda9cc9ffea07fb4e4
Link(s) to Main File(s)
Access Level
Open Access
Export
Marked PublicationsOpen Data ISTA Research Explorer
Sources
arXiv 2111.03165