1,988 Hits in 4.4 sec

A High Probability Safety Guarantee for Shifted Neural Network Surrogates

Melanie Ducoffe, Sébastien Gerchinovitz, Jayant Sen Gupta
2020 AAAI Conference on Artificial Intelligence  
deviation inequalities to estimate the probability of under-estimating a reference model with a surrogate model. • We show how to shift a surrogate to guarantee safeness with high probability. • Since  ...  We also derive a new loss function suited for shifted surrogates and study the influence of the different confidence parameters on the trade-off between the safety and accuracy of the surrogate models.  ...  Acknowledgements This project received funding from the French "Investing for the Future -PIA3" program within the Artificial and Natural Intelligence Toulouse Institute (ANITI).  ... 
dblp:conf/aaai/DucoffeGG20 fatcat:vzfopagc3jhfjcdu7pa6fg2nne

Distributionally Robust Surrogate Optimal Control for High-Dimensional Systems [article]

Aaron Kandel, Saehong Park, Scott Moura
2022 arXiv   pre-print
This paper presents a novel methodology for tractably solving optimal control and offline reinforcement learning problems for high-dimensional systems.  ...  This work is motivated by the ongoing challenges of safety, computation, and optimality in high-dimensional optimal control. We address these key questions with the following approach.  ...  For high-dimensional, large-scale nonlinear dynamical systems, the probability of this occurring is significant. Thus, safety must be guaranteed with respect to such OOD experience.  ... 
arXiv:2105.10070v2 fatcat:cx5bttn4afgitoh77juq4z4any

Perspectives on the System-level Design of a Safe Autonomous Driving Stack [article]

Majd Hawasly, Jonathan Sadeghi, Morris Antonello, Stefano V. Albrecht, John Redford, Subramanian Ramamoorthy
2022 arXiv   pre-print
This motivates going beyond extrinsic metrics such as miles between disengagement, and calls for approaches that embody safety by design.  ...  includes safe-by-design planning, interpretable as well as verifiable prediction, and modelling of perception errors to enable effective sim-to-real and real-to-sim transfer within the testing pipeline of a  ...  To address this issue, we 'distil' the performance of the 2s-OPT planner offline with a deep neural network, such that the network learns to produce similar outputs to the two-stage optimiser for the same  ... 
arXiv:2208.00096v1 fatcat:77uolkow6fcptijbf2v6v43hia

Application of Artificial Neural Networks to Chemical and Process Engineering [chapter]

Fabio Machado Cavalcanti, Camila Emilia Kozonoe, Kelvin André Pacheco, Rita Maria de Brito Alves
2021 Artificial Neural Networks and Deep Learning - Applications and Perspective [Working Title]  
The accelerated use of Artificial Neural Networks (ANNs) in Chemical and Process Engineering has drawn the attention of scientific and industrial communities, mainly due to the Big Data boom related to  ...  This empirical method can widely replace traditional complex phenomenological models based on nonlinear conservation equations, leading to a smaller computational effort – a very peculiar feature for its  ...  In addition, the authors acknowledge the financial support provided by FAPESP for doctoral scholarships (Grant 2017/11940-5 and 2017/26683-8).  ... 
doi:10.5772/intechopen.96641 fatcat:44fhkgo6anddbimc4tdsisquly

Constrained Model-Free Reinforcement Learning for Process Optimization [article]

Elton Pan, Panagiotis Petsagkourakis, Max Mowbray, Dongda Zhang, Antonio del Rio-Chanona
2021 arXiv   pre-print
We propose an 'oracle'-assisted constrained Q-learning algorithm that guarantees the satisfaction of joint chance constraints with a high probability, which is crucial for safety critical tasks.  ...  This results in a general methodology that can be imbued into approximate dynamic programming-based algorithms to ensure constraint satisfaction with high probability.  ...  In this work, we propose a Q-learning method, which guarantees constraint satisfaction with high probability.  ... 
arXiv:2011.07925v2 fatcat:4eue7eakvvhwzm2u3kuigyrh6q

Sample-Efficient Safety Assurances using Conformal Prediction [article]

Rachel Luo, Shengjia Zhao, Jonathan Kuck, Boris Ivanovic, Silvio Savarese, Edward Schmerling, Marco Pavone
2022 arXiv   pre-print
We apply our framework to a driver warning system and a robotic grasping application, and empirically demonstrate guaranteed false negative rate while also observing low false detection (positive) rate  ...  To reliably improve safety, these warning systems should have a provable false negative rate; i.e. of the situations that are unsafe, fewer than ϵ will occur without an alert.  ...  -we guarantee the failure probability for a sequence of i.i.d. failure events. -Conformal prediction guarantee.  ... 
arXiv:2109.14082v3 fatcat:vvpnzknpgvdflaohi2u3lgbngy

SmOOD: Smoothness-based Out-of-Distribution Detection Approach for Surrogate Neural Networks in Aircraft Design [article]

Houssem Ben Braiek, Ali Tfaily, Foutse Khomh, Thomas Reid, Ciro Guida
2022 arXiv   pre-print
Feedforward neural networks (FNNs) can capture highly nonlinear input-output mappings, yielding efficient surrogates for aircraft performance factors.  ...  Hybrid surrogate optimization maintains high results quality while providing rapid design assessments when both the surrogate model and the switch mechanism for eventually transitioning to the HF model  ...  We also acknowledge the support of the Canadian Institute for Advanced Research (CIFAR).  ... 
arXiv:2209.03438v1 fatcat:64rnkgtp4jbsjh5ciknw4xpvza

Probabilistic Surrogate Networks for Simulators with Unbounded Randomness [article]

Andreas Munk, Berend Zwartsenberg, Adam Ścibior, Atılım Güneş Baydin, Andrew Stewart, Goran Fernlund, Anoush Poursartip, Frank Wood
2022 arXiv   pre-print
We present a framework for automatically structuring and training fast, approximate, deep neural surrogates of stochastic simulators.  ...  We then proceed with an example that shows our surrogates are able to accurately model a complex structure like an unbounded stack in a program synthesis example.  ...  Consider a probability measure P ζ at characterized by a neural network ζ at ∈ G according to Eq. ( 7 ).  ... 
arXiv:1910.11950v2 fatcat:6ghgkxl7efgypbcthwhcvzar5q

Neural Bridge Sampling for Evaluating Safety-Critical Autonomous Systems [article]

Aman Sinha, Matthew O'Kelly, Russ Tedrake, John Duchi
2021 arXiv   pre-print
In this work, we employ a probabilistic approach to safety evaluation in simulation, where we are concerned with computing the probability of dangerous events.  ...  We provide rigorous guarantees for the performance of our method in terms of both statistical and computational efficiency.  ...  Conclusion There is a growing need for rigorous evaluation of safety-critical systems which contain components without formal guarantees (e.g. deep neural networks).  ... 
arXiv:2008.10581v3 fatcat:auitkxy2gfbadhgjg25kf4v2si

Inspect, Understand, Overcome: A Survey of Practical Methods for AI Safety [article]

Sebastian Houben, Stephanie Abrecht, Maram Akila, Andreas Bär, Felix Brockherde, Patrick Feifel, Tim Fingscheidt, Sujan Sai Gannamaneni, Seyed Eghbal Ghobadi, Ahmed Hammam, Anselm Haselhoff, Felix Hauser (+29 others)
2021 arXiv   pre-print
The use of deep neural networks (DNNs) in safety-critical applications like mobile health and autonomous driving is challenging due to numerous model-inherent shortcomings.  ...  Cyber-physical systems employing DNNs are therefore likely to suffer from safety concerns. In recent years, a zoo of state-of-the-art techniques aiming to address these safety concerns has emerged.  ...  Acknowledgment The research leading to these results is funded by the German Federal Ministry for Economic Affairs and Energy within the project "KI Absicherung -Safe AI for Automated Driving".  ... 
arXiv:2104.14235v1 fatcat:f6sj3v2brza7thyzw7b7fkpo2m

Agile Autonomous Driving using End-to-End Deep Imitation Learning

Yunpeng Pan, Ching-An Cheng, Kamil Saigol, Keuntaek Lee, Xinyan Yan, Evangelos Theodorou, Byron Boots
2018 Robotics: Science and Systems XIV  
By imitating a model predictive controller equipped with advanced sensors, we train a deep neural network control policy to map raw, high-dimensional observations to continuous steering and throttle commands  ...  We present an end-to-end imitation learning system for agile, off-road autonomous driving using only low-cost on-board sensors.  ...  CONCLUSION We introduce an end-to-end system to learn a deep neural network control policy for high-speed driving that maps raw on-board observations to steering and throttle commands by mimicking a model  ... 
doi:10.15607/rss.2018.xiv.056 dblp:conf/rss/PanCSLYTB18 fatcat:p3wthkfno5gdrammgarr32ujbq

Robustness of 3D Deep Learning in an Adversarial Setting [article]

Matthew Wicker, Marta Kwiatkowska
2019 arXiv   pre-print
In this work, we develop an algorithm for analysis of pointwise robustness of neural networks that operate on 3D data.  ...  The lack of comprehensive analysis makes it difficult to justify deployment of 3D deep learning models in real-world, safety-critical applications.  ...  FCN where FCN stands for fully connected network and refers to a neural network with potentially several layers of neurons which are fully connected.  ... 
arXiv:1904.00923v1 fatcat:m5atqg4vybgqtnhzhc5vvxczx4

Robustness of 3D Deep Learning in an Adversarial Setting

Matthew Wicker, Marta Kwiatkowska
2019 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)  
In this work, we develop an algorithm for analysis of pointwise robustness of neural networks that operate on 3D data.  ...  The lack of comprehensive analysis makes it difficult to justify deployment of 3D deep learning models in real-world, safety-critical applications.  ...  FCN where FCN stands for fully connected network and refers to a neural network with potentially several layers of neurons which are fully connected.  ... 
doi:10.1109/cvpr.2019.01204 dblp:conf/cvpr/WickerK19 fatcat:pthkgbyldjgcpe5zg7woni7msa

How to Certify Machine Learning Based Safety-critical Systems? A Systematic Literature Review [article]

Florian Tambon, Gabriel Laberge, Le An, Amin Nikanjam, Paulina Stevia Nouwou Mindom, Yann Pequignot, Foutse Khomh, Giulio Antoniol, Ettore Merlo, François Laviolette
2021 arXiv   pre-print
However, including it in so-called 'safety-critical' systems such as automotive or aeronautic has proven to be very challenging, since the shift in paradigm that ML brings completely changes traditional  ...  the question 'How to Certify Machine Learning Based Safety-critical Systems?'.  ...  currently plague modern Neural Networks.  ... 
arXiv:2107.12045v3 fatcat:43vqxywawbeflhs6ehzovvsevm

Deep surrogate accelerated delayed-acceptance HMC for Bayesian inference of spatio-temporal heat fluxes in rotating disc systems [article]

Teo Deveney, Eike Mueller, Tony Shardlow
2022 arXiv   pre-print
We introduce a training scheme that uses temperature data to adaptively train a neural-network surrogate to simulate the parametric forward model.  ...  Our approach achieves fast mixing in high-dimensional parameter spaces, whilst retaining the convergence guarantees of a traditional PDE solver, and without the burden of evaluating this solver for proposals  ...  Acknowledgements We thank our colleague Hui Tang from the Department of Mechanical Engineering at the University of Bath for their helpful discussions regarding the context of this work.  ... 
arXiv:2204.02272v1 fatcat:24bxp5ppsrfbffxaynaraq3ryi
« Previous Showing results 1 — 15 out of 1,988 results