ACTA AERONAUTICAET ASTRONAUTICA SINICA >
Safety assessment for airborne CNN classifier based on conditional Gaussian PAC-Bayes
Received date: 2024-06-14
Revised date: 2024-07-03
Accepted date: 2024-09-05
Online published: 2024-09-10
Supported by
National Key Research and Development Program of China(2022YFB3904300)
To address the airworthiness safety challenges caused by inherent uncertainty outputs of machine learning technology in airborne systems， a system safety assessment method based on the generalization theory is proposed for CNN classification under the framework of SAE ARP4761standards. First， based on the PAC-Bayes theory， the training method is improved through conditional gaussian process to optimize the generalization bound and obtain a quantified representation of the uncertainty of the CNN model. Second， an integration method for software uncertainty and hardware reliability based on generalization bound confidence is proposed to obtain comprehensive failure basic data of CNN components， supporting quantitative safety assessment of the aircraft/system. Finally， taking the airborne GNSS interference signal recognition module based on CNN as a case， the proposed method is shown to be effective in safety assessment， and is also experimentally verified that the generalization boundary based on conditional gaussian process has a tighter computational boundary than that of ordinary PAC-Bayes and Vapnik-Chervonenkis dimensions.
Zan MA , Jie BAI , Yong CHEN , Ruihua LIU , Yanting ZHANG . Safety assessment for airborne CNN classifier based on conditional Gaussian PAC-Bayes[J]. ACTA AERONAUTICAET ASTRONAUTICA SINICA, 2025 , 46(4) : 330824 -330824 . DOI: 10.7527/S1000-6893.2024.30824
| 1 | 刘嘉琛， 董磊， 陈曦， 等. 基于改进STPA-DEMATEL的智能航电系统致因要素分析［J］. 系统工程与电子技术， 2024， 46（6）： 2023-2033. |
| LIU J C， DONG L， CHEN X， et al. Analysis of causal factors of intelligent avionics system based on improved STPA-DEMTEL［J］. Systems Engineering and Electronics， 2024， 46（6）： 2023-2033 （in Chinese）. | |
| 2 | PATRICK K Y. Artificial intelligence roadmap： A human-centric approach to AI in aviation（version1.0）［R］. Cologne： EASA， 2020. |
| 3 | PATRICK K Y. Artificial intelligence roadmap： A human-centric approach to AI in aviation（version2.0）［R］. Cologne： EASA， 2023. |
| 4 | JEAN C， XAVIER H， GEORGES R， et al. Concepts of design assurance for neural networks ［R］. Cologne： EASA， 2020. |
| 5 | TORENS C， DURAK U， DAUER J C. Guidelines and regulatory framework for machine learning in aviation［C］∥Proceedings of the AIAA SCITECH 2022 Forum. Reston： AIAA， 2022. |
| 6 | JOHN D， ERIC P，BOB M， et al. Guidelines for development of civil aircraft and systems： ARP4754A ［S］. Warrendale： SAE International， 2011. |
| 7 | JOHN D， LARRY L， MICHAEL B， et al. Guidelines and methods for conducting the safety assessment process on civil airborne systems and equipment： ARP4761 ［S］. Warrendale： SAE International， 1996. |
| 8 | ROGER S， DANIEL H， MICHAEL D， et al. Software considerations in airborne systems and equipment certification： DO-178 ［S］. Washington D.C： RTCA， 2011. |
| 9 | ROBERT C， LEE J， AMAUD D， et al. Design assurance guidance for airborne electronic hardware： DO-254 ［S］. Washington D.C： RTCA， 2000. |
| 10 | JAKOB G. A survey of uncertainty in deep neural networks［DB/OL］. arXiv preprint： 2107.0334， 2022. |
| 11 | VAPNIK V N， CHERVONENKIS A Y. On the uniform convergence of relative frequencies of events to their probabilities［J］. Theory of Probability & Its Applications， 1971， 16（2）： 264-280. |
| 12 | NATARAJAN B K. On learning sets and functions［J］. Machine Learning， 1989， 4（1）： 67-97. |
| 13 | BEN-DAVID S， CESA-BIANCHI N， HAUSSLER D， et al. Characterizations of learnability for classes of｛0， ..， n｝-valued functions［J］. Journal of Computer and System Sciences， 1995， 50： 74-86. |
| 14 | MCALLESTER D A. PAC-Bayesian model averaging［C］??∥Proceedings of the twelfth annual conference on Computational learning theory. New York： ACM， 1999. |
| 15 | BéGIN L， GERMAIN P， LAVIOLETTE F， et al. PAC-Bayesian theory for transductive learning［C］?∥Proceedings of the Seventeenth International Conference on Artificial Intelligence and Statistics. Reykjavik： PMLR， 2014. |
| 16 | 李旭燕， 任晓. 主观解释的新发展： 主体交互解释［J］. 心智与计算， 2007， 1（3）： 293-299. |
| LI X Y， REN X. The new development of subjective interpretation： Subject interaction interpretation［J］. Mind and Computing， 2007， 1（3）： 293-299 （in Chinese）. | |
| 17 | WASHINGTON A， CLOTHIER R A， WILLIAMS B P. A Bayesian approach to system safety assessment and compliance assessment for unmanned aircraft systems［J］. Journal of Air Transport Management， 2017， 62： 18-33. |
| 18 | WASHINGTON A， CLOTHIER R， NEOGI N， et al. Adoption of a Bayesian belief network for the system safety assessment of remotely piloted aircraft systems［J］. Safety Science， 2019， 118： 654-673. |
| 19 | 王海朋， 段富海. 复杂不确定系统可靠性分析的贝叶斯网络方法［J］. 兵工学报， 2020， 41（1）： 171-182. |
| WANG H P， DUAN F H. Bayesian network method for reliability analysis of complex uncertainty systems［J］. Acta Armamentarii， 2020， 41（1）： 171-182 （in Chinese）. | |
| 20 | LANGFORD J. Tutorial on practical prediction theory for classification［J］. Journal of Machine Learning Research， 2005， 6： 273-306. |
| 21 | MCALLESTER D A. PAC-Bayesian stochastic model selection［J］. Machine Learning， 2003， 51（1）： 5-21. |
| 22 | BIGGS F， GUEDJ B. Differentiable PAC-bayes objectives with partially aggregated neural networks［J］. Entropy， 2021， 23（10）： 1280. |
| 23 | CLERICO E， DELIGIANNIDIS G， DOUCET A. Wide stochastic networks： Gaussian limit and PAC-Bayesian training［DB/OL］. arXiv preprint： 2016. 09798， 2021. |
| 24 | XU H， CHENG Y F， LIANG J D， et al. A jamming recognition algorithm based on deep neural network in satellite navigation system［M］∥Lecture Notes in Electrical Engineering. Singapore： Springer Singapore， 2020： 701-711. |
/
| 〈 |
|
〉 |
CJA