Improving Robustness of Deep Neural Networks for Aerial Navigation by Incorporating Input Uncertainty

Fabio Arnez; Huascar Espinoza; Ansgar Radermacher; François Terrier

doi:10.1007/978-3-030-83906-2_17

Improving Robustness of Deep Neural Networks for Aerial Navigation by Incorporating Input Uncertainty

Fabio Arnez^*
, Huascar Espinoza
, Ansgar Radermacher
, François Terrier

^*Corresponding author for this work

Université Paris-Saclay

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

4 Citations (Scopus)

Abstract

Uncertainty quantification methods are required in autonomous systems that include deep learning (DL) components to assess the confidence of their estimations. However, to successfully deploy DL components in safety-critical autonomous systems, they should also handle uncertainty at the input rather than only at the output of the DL components. Considering a probability distribution in the input enables the propagation of uncertainty through different components to provide a representative measure of the overall system uncertainty. In this position paper, we propose a method to account for uncertainty at the input of Bayesian Deep Learning control policies for Aerial Navigation. Our early experiments show that the proposed method improves the robustness of the navigation policy in Out-of-Distribution (OoD) scenarios.

Original language	English
Title of host publication	Computer Safety, Reliability, and Security. SAFECOMP 2021 Workshops - DECSoS, MAPSOD, DepDevOps, USDAI, and WAISE, Proceedings
Editors	Ibrahim Habli, Mark Sujan, Simos Gerasimou, Erwin Schoitsch, Friedemann Bitsch
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	219-225
Number of pages	7
ISBN (Print)	9783030839055
DOIs	https://doi.org/10.1007/978-3-030-83906-2_17
Publication status	Published - 2021
Externally published	Yes
Event	40th International Conference on Computer Safety, Reliability and Security, SAFECOMP 2021 held in conjunction with Workshops on DECSoS, MAPSOD, DepDevOps, USDAI and WAISE 2021 - Virtual, Online Duration: 7 Sept 2021 → 10 Sept 2021

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	12853 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	40th International Conference on Computer Safety, Reliability and Security, SAFECOMP 2021 held in conjunction with Workshops on DECSoS, MAPSOD, DepDevOps, USDAI and WAISE 2021
City	Virtual, Online
Period	7/09/21 → 10/09/21

Keywords

AI safety
Autonomous systems
Uncertainty propagation

Access to Document

10.1007/978-3-030-83906-2_17

Cite this

Arnez, F., Espinoza, H., Radermacher, A., & Terrier, F. (2021). Improving Robustness of Deep Neural Networks for Aerial Navigation by Incorporating Input Uncertainty. In I. Habli, M. Sujan, S. Gerasimou, E. Schoitsch, & F. Bitsch (Eds.), Computer Safety, Reliability, and Security. SAFECOMP 2021 Workshops - DECSoS, MAPSOD, DepDevOps, USDAI, and WAISE, Proceedings (pp. 219-225). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12853 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-030-83906-2_17

Arnez, Fabio ; Espinoza, Huascar ; Radermacher, Ansgar et al. / Improving Robustness of Deep Neural Networks for Aerial Navigation by Incorporating Input Uncertainty. Computer Safety, Reliability, and Security. SAFECOMP 2021 Workshops - DECSoS, MAPSOD, DepDevOps, USDAI, and WAISE, Proceedings. editor / Ibrahim Habli ; Mark Sujan ; Simos Gerasimou ; Erwin Schoitsch ; Friedemann Bitsch. Springer Science and Business Media Deutschland GmbH, 2021. pp. 219-225 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{e04c67ae135c46a0acc8d6779a31a650,

title = "Improving Robustness of Deep Neural Networks for Aerial Navigation by Incorporating Input Uncertainty",

abstract = "Uncertainty quantification methods are required in autonomous systems that include deep learning (DL) components to assess the confidence of their estimations. However, to successfully deploy DL components in safety-critical autonomous systems, they should also handle uncertainty at the input rather than only at the output of the DL components. Considering a probability distribution in the input enables the propagation of uncertainty through different components to provide a representative measure of the overall system uncertainty. In this position paper, we propose a method to account for uncertainty at the input of Bayesian Deep Learning control policies for Aerial Navigation. Our early experiments show that the proposed method improves the robustness of the navigation policy in Out-of-Distribution (OoD) scenarios.",

keywords = "AI safety, Autonomous systems, Uncertainty propagation",

author = "Fabio Arnez and Huascar Espinoza and Ansgar Radermacher and Fran{\c c}ois Terrier",

note = "Publisher Copyright: {\textcopyright} 2021, Springer Nature Switzerland AG.; 40th International Conference on Computer Safety, Reliability and Security, SAFECOMP 2021 held in conjunction with Workshops on DECSoS, MAPSOD, DepDevOps, USDAI and WAISE 2021 ; Conference date: 07-09-2021 Through 10-09-2021",

year = "2021",

doi = "10.1007/978-3-030-83906-2\_17",

language = "English",

isbn = "9783030839055",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "219--225",

editor = "Ibrahim Habli and Mark Sujan and Simos Gerasimou and Erwin Schoitsch and Friedemann Bitsch",

booktitle = "Computer Safety, Reliability, and Security. SAFECOMP 2021 Workshops - DECSoS, MAPSOD, DepDevOps, USDAI, and WAISE, Proceedings",

address = "Germany",

}

Arnez, F, Espinoza, H, Radermacher, A & Terrier, F 2021, Improving Robustness of Deep Neural Networks for Aerial Navigation by Incorporating Input Uncertainty. in I Habli, M Sujan, S Gerasimou, E Schoitsch & F Bitsch (eds), Computer Safety, Reliability, and Security. SAFECOMP 2021 Workshops - DECSoS, MAPSOD, DepDevOps, USDAI, and WAISE, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 12853 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 219-225, 40th International Conference on Computer Safety, Reliability and Security, SAFECOMP 2021 held in conjunction with Workshops on DECSoS, MAPSOD, DepDevOps, USDAI and WAISE 2021, Virtual, Online, 7/09/21. https://doi.org/10.1007/978-3-030-83906-2_17

Improving Robustness of Deep Neural Networks for Aerial Navigation by Incorporating Input Uncertainty. / Arnez, Fabio; Espinoza, Huascar; Radermacher, Ansgar et al.
Computer Safety, Reliability, and Security. SAFECOMP 2021 Workshops - DECSoS, MAPSOD, DepDevOps, USDAI, and WAISE, Proceedings. ed. / Ibrahim Habli; Mark Sujan; Simos Gerasimou; Erwin Schoitsch; Friedemann Bitsch. Springer Science and Business Media Deutschland GmbH, 2021. p. 219-225 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12853 LNCS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Improving Robustness of Deep Neural Networks for Aerial Navigation by Incorporating Input Uncertainty

AU - Arnez, Fabio

AU - Espinoza, Huascar

AU - Radermacher, Ansgar

AU - Terrier, François

PY - 2021

Y1 - 2021

N2 - Uncertainty quantification methods are required in autonomous systems that include deep learning (DL) components to assess the confidence of their estimations. However, to successfully deploy DL components in safety-critical autonomous systems, they should also handle uncertainty at the input rather than only at the output of the DL components. Considering a probability distribution in the input enables the propagation of uncertainty through different components to provide a representative measure of the overall system uncertainty. In this position paper, we propose a method to account for uncertainty at the input of Bayesian Deep Learning control policies for Aerial Navigation. Our early experiments show that the proposed method improves the robustness of the navigation policy in Out-of-Distribution (OoD) scenarios.

AB - Uncertainty quantification methods are required in autonomous systems that include deep learning (DL) components to assess the confidence of their estimations. However, to successfully deploy DL components in safety-critical autonomous systems, they should also handle uncertainty at the input rather than only at the output of the DL components. Considering a probability distribution in the input enables the propagation of uncertainty through different components to provide a representative measure of the overall system uncertainty. In this position paper, we propose a method to account for uncertainty at the input of Bayesian Deep Learning control policies for Aerial Navigation. Our early experiments show that the proposed method improves the robustness of the navigation policy in Out-of-Distribution (OoD) scenarios.

KW - AI safety

KW - Autonomous systems

KW - Uncertainty propagation

UR - https://www.scopus.com/pages/publications/85115176040

U2 - 10.1007/978-3-030-83906-2_17

DO - 10.1007/978-3-030-83906-2_17

M3 - Conference contribution

AN - SCOPUS:85115176040

SN - 9783030839055

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 219

EP - 225

BT - Computer Safety, Reliability, and Security. SAFECOMP 2021 Workshops - DECSoS, MAPSOD, DepDevOps, USDAI, and WAISE, Proceedings

A2 - Habli, Ibrahim

A2 - Sujan, Mark

A2 - Gerasimou, Simos

A2 - Schoitsch, Erwin

A2 - Bitsch, Friedemann

PB - Springer Science and Business Media Deutschland GmbH

T2 - 40th International Conference on Computer Safety, Reliability and Security, SAFECOMP 2021 held in conjunction with Workshops on DECSoS, MAPSOD, DepDevOps, USDAI and WAISE 2021

Y2 - 7 September 2021 through 10 September 2021

ER -

Arnez F, Espinoza H, Radermacher A, Terrier F. Improving Robustness of Deep Neural Networks for Aerial Navigation by Incorporating Input Uncertainty. In Habli I, Sujan M, Gerasimou S, Schoitsch E, Bitsch F, editors, Computer Safety, Reliability, and Security. SAFECOMP 2021 Workshops - DECSoS, MAPSOD, DepDevOps, USDAI, and WAISE, Proceedings. Springer Science and Business Media Deutschland GmbH. 2021. p. 219-225. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-83906-2_17

Improving Robustness of Deep Neural Networks for Aerial Navigation by Incorporating Input Uncertainty

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this