arxiv: 2604.26411 · v1 · submitted 2026-04-29 · 💻 cs.LG

Recognition: unknown

Unifying Runtime Monitoring Approaches for Safety-Critical Machine Learning: Application to Vision-Based Landing

Mathieu Dario , Florent Chenevier , K\'evin Delmas , Joris Guerin , J\'er\'emie Guiochet

Authors on Pith no claims yet

Pith reviewed 2026-05-07 11:09 UTC · model grok-4.3

classification 💻 cs.LG

keywords runtime monitoringsafety-critical machine learningoperational design domainout-of-distribution detectionout-of-model-scopevision-based landingrunway detectionunified framework

0 comments

The pith

Runtime monitoring for safety-critical machine learning can be unified by dividing approaches into three categories that check operating conditions, input distribution, or model behavior.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Current research on runtime monitoring for ML in safety-critical settings is scattered across communities with incompatible methods. The paper introduces a framework that sorts these approaches into three groups: Operational Design Domain monitoring to verify the system stays in its intended operating environment, Out-of-Distribution monitoring to flag inputs far from the training data, and Out-of-Model-Scope monitoring to identify when the model's own behavior becomes unreliable. This division makes it easier to build sets of monitors that cover different failure modes without redundancy. It also creates a common set of safety-focused metrics so researchers can directly compare how well different monitors work. The authors test the idea on a runway detection task for aircraft landing to show how the categories help in practice.

Core claim

The central discovery is that categorizing runtime monitoring into Operational Design Domain (ODD) monitoring for compliance with expected conditions, Out-of-Distribution (OOD) monitoring for inputs outside the training distribution, and Out-of-Model-Scope (OMS) monitoring for anomalous internal model states or outputs unifies disparate approaches. This categorization facilitates the design of complementary monitoring activities and allows evaluation using shared safety-oriented metrics, as demonstrated in an experiment on vision-based runway detection during landing.

What carries the argument

The three distinct monitoring categories—Operational Design Domain (ODD), Out-of-Distribution (OOD), and Out-of-Model-Scope (OMS)—that classify approaches according to whether they check operating conditions, input distribution, or model behavior.

If this is right

Monitoring activities can be designed using monitors from complementary categories to achieve broader coverage.
Different monitoring methods can be evaluated and compared using the same safety-oriented metrics.
The framework supports applications in safety-critical domains such as aeronautics, specifically vision-based landing.
Fragmented research from different communities can be integrated under a single categorization scheme.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

This unification could enable the creation of standardized test suites for ML monitors across industries.
Extending the categories to other safety-critical ML uses like medical diagnosis might highlight missing monitor types.
Future work could develop hybrid monitors that combine elements from multiple categories for improved performance.

Load-bearing premise

The three proposed categories are distinct and comprehensive enough to cover existing runtime monitoring approaches without major overlap or omission, and the runway detection experiment demonstrates practical benefits of the framework.

What would settle it

A runtime monitoring technique for safety-critical ML that cannot be placed in any of the three categories without forcing overlap or omission, or an evaluation in the runway detection task where categorized monitors show no advantage in safety metrics over existing methods.

Figures

Figures reproduced from arXiv: 2604.26411 by Florent Chenevier, J\'er\'emie Guiochet, Joris Guerin, K\'evin Delmas, Mathieu Dario.

**Figure 1.** Figure 1: Illustration of the Vision-Based Runway Detection (VBRD) task. view at source ↗

**Figure 2.** Figure 2: Decomposition of the ML input space and definition of data domains and data threats. view at source ↗

**Figure 3.** Figure 3: Experimental setup – serial monitoring architecture. view at source ↗

**Figure 4.** Figure 4: Corruptions applied to the image in Figure view at source ↗

**Figure 5.** Figure 5: Visualisation of the cumulated Safety Gain ( view at source ↗

read the original abstract

Runtime monitoring is essential to ensure the safety of ML applications in safety-critical domains. However, current research is fragmented, with independent methods emerging from different communities. In this paper, we propose a unified framework categorising runtime monitoring approaches into three distinct types: Operational Design Domain (ODD) monitoring, which ensures compliance with expected operating conditions; Out-of-Distribution (OOD) monitoring, which rejects inputs that deviate from the training data; and Out-of-Model-Scope (OMS) monitoring, which detects anomalous model behaviour based its internal states or outputs. We demonstrate the benefits of this categorization with a dedicated experiment on an aeronautical safety-critical application: runway detection during landing. This framework facilitates design of monitoring activities, with complementary categories of monitors, and enables evaluation and comparison of different monitors using common, safety-oriented metrics.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper gives a clean three-way taxonomy for runtime monitoring in safety-critical ML and illustrates it on runway detection, but the categories stay fuzzy and the experiment stays illustrative.

read the letter

This paper proposes a three-way taxonomy for runtime monitoring in safety-critical machine learning and tests it on a vision-based landing task, but the categories need tighter definitions to deliver on the unification promise. The new part is the explicit split into ODD monitoring (staying inside the operational design domain), OOD monitoring (catching inputs far from training data), and OMS monitoring (spotting weird behavior inside the model or its outputs). They apply this to runway detection during aircraft landing, which is a good concrete case from a regulated domain. What works is the high-level organization. It gives designers a checklist: cover the operating envelope, reject bad inputs, and watch for model failures. The aviation example makes the abstract categories feel practical. The weak points are the lack of crisp boundaries and evidence. The paper defines the three types intuitively but does not spell out how to classify a monitor that looks at both input statistics and model logits, for example. There is also no systematic review showing that most existing methods fit neatly into one bin without overlap. The experiment demonstrates one monitor per category but stops short of quantitative results on how the framework improves safety or allows fair comparison. That leaves the main claim—that this enables better design and evaluation—more as a suggestion than a demonstrated result. This work is for engineers and researchers building monitors for ML in aviation or similar safety-critical fields. Someone looking for a starting structure to compare monitors will find it useful, though they will need to do extra work on the edges. It is worth sending to peer review. The idea is clear and the domain is important, so referees can help tighten the taxonomy and ask for more data on the case study.

Referee Report

3 major / 2 minor

Summary. The paper proposes a unified framework for runtime monitoring of safety-critical ML systems by categorizing approaches into three types: Operational Design Domain (ODD) monitoring to ensure compliance with expected operating conditions, Out-of-Distribution (OOD) monitoring to reject inputs deviating from training data, and Out-of-Model-Scope (OMS) monitoring to detect anomalous model behavior from internal states or outputs. It illustrates the framework via a runway detection experiment in a vision-based landing application, claiming that the categorization enables complementary monitor design and evaluation/comparison of monitors using shared safety-oriented metrics.

Significance. If the taxonomy proves comprehensive and the common metrics enable rigorous cross-monitor comparisons, the work could reduce fragmentation across communities (e.g., OOD detection, anomaly detection, and operational monitoring) and support more systematic safety case construction for ML in domains like aviation. The runway-detection case study provides a concrete aeronautical example, which is a strength for applicability, but the absence of quantitative results, error analysis, or a literature mapping table limits the demonstrated impact on evaluation.

major comments (3)

[§3] §3 (Proposed Framework): The definitions of ODD, OOD, and OMS monitoring are given intuitively but without explicit decision procedures or boundary conditions for borderline cases (e.g., an input-distribution monitor that also inspects model logits or outputs). This directly undermines the claim that the categories are 'distinct' and 'complementary' and that they enable unambiguous design and comparison.
[§4] §4 (Case Study / Experiment): The runway detection demonstration assigns one monitor to each category but provides no systematic mapping of prior runtime-monitoring papers to the three bins, nor does it report quantitative results, safety-oriented metrics, or comparative evaluation across monitors. Without this, the experiment illustrates rather than validates the unification and common-metric benefits asserted in the abstract and §1.
[§2] §2 (Related Work): No table or structured review maps existing methods (e.g., from OOD detection literature or runtime verification) onto the ODD/OOD/OMS taxonomy. This omission is load-bearing for the 'unifying' claim, as overlap or omission risks cannot be assessed.

minor comments (2)

[§3] Notation for the three categories is introduced without a summary table; adding one early in §3 would improve readability.
[Abstract and §4] The abstract states the experiment demonstrates 'benefits' but the text provides only qualitative description; clarify whether any numerical safety metrics (e.g., false-positive rates under ODD violation) appear in §4.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments, which help us improve the clarity and rigor of our proposed framework. We respond to each major comment below and describe the corresponding revisions to the manuscript.

read point-by-point responses

Referee: [§3] §3 (Proposed Framework): The definitions of ODD, OOD, and OMS monitoring are given intuitively but without explicit decision procedures or boundary conditions for borderline cases (e.g., an input-distribution monitor that also inspects model logits or outputs). This directly undermines the claim that the categories are 'distinct' and 'complementary' and that they enable unambiguous design and comparison.

Authors: We agree that the intuitive definitions in Section 3 would benefit from more explicit decision procedures to handle borderline cases and to rigorously support the claims of distinctness and complementarity. In the revised manuscript, we will augment §3 with a decision flowchart or pseudocode outlining the classification criteria for each category, along with concrete examples of monitors that might straddle boundaries (such as those examining both input distributions and model outputs). This addition will clarify how the categories remain distinct while allowing for complementary use in safety-critical applications. revision: yes
Referee: [§4] §4 (Case Study / Experiment): The runway detection demonstration assigns one monitor to each category but provides no systematic mapping of prior runtime-monitoring papers to the three bins, nor does it report quantitative results, safety-oriented metrics, or comparative evaluation across monitors. Without this, the experiment illustrates rather than validates the unification and common-metric benefits asserted in the abstract and §1.

Authors: The experiment in §4 serves primarily to demonstrate the application of the proposed framework in a concrete aeronautical scenario. We acknowledge that it does not include a full comparative evaluation or quantitative safety metrics. To address this, we will expand §4 to include quantitative results using safety-oriented metrics such as detection rates, false alarm rates, and coverage of the operational envelope. We will also provide a comparative analysis of the three monitors. However, due to the illustrative nature and computational constraints of the simulation environment, an exhaustive validation across all prior methods is not feasible; instead, we will focus on the assigned monitors and discuss how the metrics enable future comparisons. revision: partial
Referee: [§2] §2 (Related Work): No table or structured review maps existing methods (e.g., from OOD detection literature or runtime verification) onto the ODD/OOD/OMS taxonomy. This omission is load-bearing for the 'unifying' claim, as overlap or omission risks cannot be assessed.

Authors: We concur that a structured mapping of the literature is essential to substantiate the unifying aspect of the framework. We will introduce a new table in §2 that categorizes representative works from OOD detection, anomaly detection, and runtime verification communities according to the ODD, OOD, and OMS types. The table will highlight potential overlaps and gaps, thereby allowing readers to assess the taxonomy's coverage and supporting the claim that the categorization facilitates systematic design and comparison. revision: yes

Circularity Check

0 steps flagged

No circularity: taxonomy built from external concepts with no self-referential derivations

full rationale

The paper advances a conceptual categorization of runtime monitoring into ODD, OOD, and OMS without equations, fitted parameters, or predictions. No step reduces a claimed result to a quantity defined by the taxonomy itself or by self-citation chains. The runway-detection experiment applies the categories illustratively rather than deriving metrics or complementarity from the framework by construction. The absence of a systematic mapping table or decision procedure is a limitation of completeness, not a circularity in the derivation chain.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 3 invented entities

The central claim rests on the domain assumption that runtime monitoring methods naturally partition into the three proposed categories and that this partition yields complementary coverage. No free parameters are introduced. The three category names function as invented organizing concepts without independent falsifiable handles outside the paper.

axioms (1)

domain assumption Runtime monitoring approaches for safety-critical ML can be partitioned into ODD, OOD, and OMS categories that are distinct, comprehensive, and complementary.
Invoked when the framework is presented as unifying and when benefits for design and comparison are claimed.

invented entities (3)

ODD monitoring no independent evidence
purpose: Ensures compliance with expected operating conditions
New label and grouping introduced to organize prior work.
OOD monitoring no independent evidence
purpose: Rejects inputs that deviate from the training data
New label and grouping introduced to organize prior work.
OMS monitoring no independent evidence
purpose: Detects anomalous model behaviour based on internal states or outputs
New label and grouping introduced to organize prior work.

pith-pipeline@v0.9.0 · 5457 in / 1430 out tokens · 75189 ms · 2026-05-07T11:09:41.064699+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

49 extracted references · 7 canonical work pages · 2 internal anchors

[1]

Concrete Problems in AI Safety

Amodei, D., Olah, C., Steinhardt, J., Christiano, P., Schulman, J., Mané, D.: Concrete problems in AI safety. arXiv preprint arXiv:1606.06565 (2016)

work page internal anchor Pith review arXiv 2016
[2]

In: Conformal and Probabilistic Prediction with Applications

Andéol, L., Fel, T., De Grancey, F., Mossina, L.: Confident object detection via conformal prediction and conformal risk control: an application to railway signaling. In: Conformal and Probabilistic Prediction with Applications. pp. 36–55. PMLR (2023)

2023
[3]

IEEE Transactions on Dependable and Secure Computing1, 11–33 (2004)

Avizienis, A., Laprie, J.C., Randell, B., Landwehr, C.: Basic Concepts and Taxonomy of Dependable and Secure Computing. IEEE Transactions on Dependable and Secure Computing1, 11–33 (2004)

2004
[4]

In: 2020 ACM/IEEE 11th International Conference on Cyber-Physical Systems (ICCPS)

Cai,F.,Koutsoukos,X.:Real-time out-of-distributiondetectioninlearning-enabledcyber-physicalsystems. In: 2020 ACM/IEEE 11th International Conference on Cyber-Physical Systems (ICCPS). pp. 174–183. IEEE (2020)

2020
[5]

Cappi, C., Cohen, N., Ducoffe, M., Gabreau, C., Gardes, L., Gauffriau, A., Ginestet, J.B., Mamalet, F., Mussot, V., Pagetti, C., et al.: How to design a dataset compliant with an ML-based system ODD? In: 12th European Congress on Embedded Real Time Software and Systems (2024)

2024
[6]

In: 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

Chen, Y., Cheng, C.H., Yan, J., Yan, R.: Monitoring object detection abnormalities via data-label and post- algorithm abstractions. In: 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). pp. 6688–6693. IEEE (2021)

2021
[7]

In: 2019 Design, Automation & Test in Europe Conference & Exhibition (DATE)

Cheng, C.H., Nührenberg, G., Yasuoka, H.: Runtime monitoring neuron activation patterns. In: 2019 Design, Automation & Test in Europe Conference & Exhibition (DATE). pp. 300–303. IEEE (2019)

2019
[8]

arXiv preprint arXiv:2306.08447 (2023)

Cheng, C.H., Wu, C., Ruess, H., Bensalem, S.: Towards Rigorous Design of OoD Detectors. arXiv preprint arXiv:2306.08447 (2023)

work page arXiv 2023
[9]

arXiv preprint arXiv:1812.02765 (2018)

Denouden, T., Salay, R., Czarnecki, K., Abdelzad, V., Phan, B., Vernekar, S.: Improving reconstruction au- toencoder out-of-distribution detection with mahalanobis distance. arXiv preprint arXiv:1812.02765 (2018)

work page arXiv 2018
[10]

Advances in neural information processing systems35, 20434–20449 (2022)

Du, X., Gozum, G., Ming, Y., Li, Y.: Siren: Shaping representations for detecting out-of-distribution objects. Advances in neural information processing systems35, 20434–20449 (2022)

2022
[11]

arXiv preprint arXiv:2202.01197 (2022)

Du, X., Wang, Z., Cai, M., Li, Y.: Vos: Learning what you don’t know by virtual outlier synthesis. arXiv preprint arXiv:2202.01197 (2022)

work page arXiv 2022
[12]

arXiv preprint arXiv:2304.09938 (2023)

Ducoffe, M., Carrere, M., Féliers, L., Gauffriau, A., Mussot, V., Pagetti, C., Sammour, T.: LARD–Landing Approach Runway Detection–Dataset for Vision Based Landing. arXiv preprint arXiv:2304.09938 (2023)

work page arXiv 2023
[13]

European Aviation Safety Agency (EASA): Artificial Intelligence Roadmap 1.0: A Human-Centric Ap- proach to AI in Aviation. Tech. rep., European Aviation Safety Agency (2020)

2020
[14]

European Aviation Safety Agency (EASA): EASA Concept Paper: Guidance for Level 1 & 2 Machine Learning Applications Issue 02. Tech. rep., European Aviation Safety Agency (2024)

2024
[15]

In: 2021 IEEE 26th Pacific Rim International Symposium on Dependable Computing (PRDC)

Ferreira, R.S., Arlat, J., Guiochet, J., Waeselynck, H.: Benchmarking safety monitors for image classi- fiers with machine learning. In: 2021 IEEE 26th Pacific Rim International Symposium on Dependable Computing (PRDC). pp. 7–16. IEEE (2021)

2021
[16]

Computational Intelligence41, e70032 (2025)

Ferreira, R.S., Guérin, J., Delmas, K., Guiochet, J., Waeselynck, H.: Safety monitoring of machine learning perception functions: a survey. Computational Intelligence41, e70032 (2025)

2025
[17]

In: international conference on machine learning

Gal, Y., Ghahramani, Z.: Dropout as a bayesian approximation: Representing model uncertainty in deep learning. In: international conference on machine learning. pp. 1050–1059. PMLR (2016)

2016
[18]

Explaining and Harnessing Adversarial Examples

Goodfellow, I.J., Shlens, J., Szegedy, C.: Explaining and harnessing adversarial examples. arXiv preprint arXiv:1412.6572 (2014)

work page internal anchor Pith review arXiv 2014
[19]

In: Proceedings of the AAAI conference on artificial intelligence

Guérin, J., Delmas, K., Ferreira, R., Guiochet, J.: Out-of-distribution detection is not all you need. In: Proceedings of the AAAI conference on artificial intelligence. vol. 37, pp. 14829–14837 (2023)

2023
[20]

In: 2022 IEEE 33rd International Symposium on Software Reliability Engineering (ISSRE)

Guerin, J., Ferreira, R.S., Delmas, K., Guiochet, J.: Unifying evaluation of machine learning safety mon- itors. In: 2022 IEEE 33rd International Symposium on Software Reliability Engineering (ISSRE). pp. 414–422. IEEE (2022)

2022
[21]

In: International Conference on Runtime Verification

Hashemi, V., Křetínsk` y, J., Rieder, S., Schön, T., Vorhoff, J.: Gaussian-based and outside-the-box runtime monitoring join forces. In: International Conference on Runtime Verification. pp. 218–228. Springer (2024)

2024
[22]

In: International Conference on Runtime Verification

He, W., Wu, C., Bensalem, S.: Box-based monitor approach for out-of-distribution detection in YOLO: an exploratory study. In: International Conference on Runtime Verification. pp. 229–239. Springer (2024)

2024
[23]

In: International Conference on Learning Representations (2019)

Hendrycks, D., Dietterich, T.: Benchmarking neural network robustness to common corruptions and per- turbations. In: International Conference on Learning Representations (2019)

2019
[24]

In: International Conference on Learning Representations (2017)

Hendrycks, D., Gimpel, K.: A baseline for detecting misclassified and out-of-distribution examples in neural networks. In: International Conference on Learning Representations (2017)

2017
[25]

In: 24th European Conference on Artificial Intelligence

Henzinger, T.A., Lukina, A., Schilling, C.: Outside the box: Abstraction-based monitoring of neural net- works. In: 24th European Conference on Artificial Intelligence. vol. 325 (2020)

2020
[26]

In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

Hsu, Y.C., Shen, Y., Jin, H., Kira, Z.: Generalized ODIN: Detecting out-of-distribution image without learning from out-of-distribution data. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 10951–10960 (2020)

2020
[27]

In: International Conference on Computer Safety, Reliability, and Security

Kaakai, F., Adibhatla, S., Pai, G., Escorihuela, E.: Data-centric operational design domain characteriza- tion for machine learning-based aeronautical products. In: International Conference on Computer Safety, Reliability, and Security. pp. 227–242. Springer (2023)

2023
[28]

In: NeurIPS MLSys Workshop

Kang, D., Raghavan, D., Bailis, P., Zaharia, M.: Model assertions for debugging machine learning. In: NeurIPS MLSys Workshop. vol. 3 (2018)

2018
[29]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

Kirchheim, K., Filax, M., Ortmeier, F.: Pytorch-ood: A library for out-of-distribution detection based on pytorch. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 4351–4360 (2022)

2022
[30]

Advances in neural information processing systems31(2018)

Lee, K., Lee, K., Lee, H., Shin, J.: A simple unified framework for detecting out-of-distribution samples and adversarial attacks. Advances in neural information processing systems31(2018)

2018
[31]

Advances in neural information processing systems33, 21464–21475 (2020)

Liu, W., Wang, X., Owens, J., Li, Y.: Energy-based out-of-distribution detection. Advances in neural information processing systems33, 21464–21475 (2020)

2020
[32]

IEEE Transactions on Systems, Man, and Cybernetics: Systems 48, 702–715 (2016)

Machin, M., Guiochet, J., Waeselynck, H., Blanquart, J.P., Roy, M., Masson, L.: SMOF: A safety monitor- ing framework for autonomous systems. IEEE Transactions on Systems, Man, and Cybernetics: Systems 48, 702–715 (2016)

2016
[33]

Digital signal processing73, 1–15 (2018)

Montavon, G., Samek, W., Müller, K.R.: Methods for interpreting and understanding deep neural networks. Digital signal processing73, 1–15 (2018)

2018
[34]

In: Proc

Ndong, J., Salamatian, K.: Signal processing-based anomaly detection techniques: a comparative analysis. In: Proc. 2011 3rd International Conference on Evolving Internet. pp. 32–39. IARIA (2011)

2011
[35]

In: Proceedings of the IEEE conference on computer vision and pattern recognition

Nguyen, A., Yosinski, J., Clune, J.: Deep neural networks are easily fooled: High confidence predictions for unrecognizable images. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 427–436 (2015)

2015
[36]

Novello, P., Prudent, Y., Friedrich, C., Pequignot, Y., Le Goff, M.: OODEEL, a simple, compact, and hackable post-hoc deep OOD detection for already trained tensorflow or pytorch image classifiers. (2023)

2023
[37]

In: 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

Rahman, Q.M., Sünderhauf, N., Dayoub, F.: Did you miss the sign? a false negative alarm system for traffic sign detectors. In: 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). pp. 3748–3753. IEEE (2019)

2019
[38]

In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision

Rahman, Q.M., Sunderhauf, N., Dayoub, F.: Per-frame map prediction for continuous performance mon- itoring of object detection during deployment. In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. pp. 152–160 (2021)

2021
[39]

SAE International: Taxonomy and definitions for terms related to driving automation systems for on-road motor vehicles. Tech. Rep. J3016, SAE International, Warrendale, PA, USA (Apr 2021)

2021
[40]

Advances in neural information processing systems34, 144–157 (2021)

Sun, Y., Guo, C., Li, Y.: React: Out-of-distribution detection with rectified activations. Advances in neural information processing systems34, 144–157 (2021)

2021
[41]

In: Interna- tional conference on machine learning

Sun, Y., Ming, Y., Zhu, X., Li, Y.: Out-of-distribution detection with deep nearest neighbors. In: Interna- tional conference on machine learning. pp. 20827–20840. PMLR (2022)

2022
[42]

CEAS Aeronautical Journal 16, 973–991 (2025)

Torens, C., Jünger, F., Schirmer, S., Nagarajan, P., Schopferer, S., Zhukov, D., Dauer, J.: Runtime monitor- ing of operational design domain to safeguard machine learning components. CEAS Aeronautical Journal 16, 973–991 (2025)

2025
[43]

Wang,H.,Li,Z.,Feng,L.,Zhang,W.:Vim:Out-of-distributionwithvirtual-logitmatching.In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 4921–4930 (2022)

2022
[44]

In: Proceedings of the ACM/IEEE 42nd International Conference on Software Engineering

Wang, H., Xu, J., Xu, C., Ma, X., Lu, J.: Dissector: Input validation for deep learning applications by crossing-layer dissection. In: Proceedings of the ACM/IEEE 42nd International Conference on Software Engineering. pp. 727–738 (2020)

2020
[45]

In: Proceedings of the IEEE/CVF international conference on computer vision

Wilson, S., Fischer, T., Dayoub, F., Miller, D., Sünderhauf, N.: Safe: Sensitivity-aware features for out- of-distribution object detection. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 23565–23576 (2023)

2023
[46]

Applied Soft Computing83, 105613 (2019)

Yahaya, S.W., Lotfi, A., Mahmud, M.: A consensus novelty detection ensemble approach for anomaly detection in activities of daily living. Applied Soft Computing83, 105613 (2019)

2019
[47]

Yang,J.,Zhou,K.,Li,Y.,Liu,Z.:Generalizedout-of-distributiondetection:Asurvey.InternationalJournal of Computer Vision132(12), 5635–5662 (2024)

2024
[48]

In: NeurIPS 2023 Workshop on Distribution Shifts: New Frontiers with Foundation Models (2023)

Zhang, J., Yang, J., Wang, P., Wang, H., Lin, Y., Zhang, H., Sun, Y., Du, X., Li, Y., Liu, Z., Chen, Y., Li, H.: OpenOOD v1.5: Enhanced Benchmark for Out-of-Distribution Detection. In: NeurIPS 2023 Workshop on Distribution Shifts: New Frontiers with Foundation Models (2023)

2023
[49]

arXiv preprint arXiv:2505.16740 (2025)

Zouzou, A., Ducoffe, M., Boumazouza, R., et al.: Robust vision-based runway detection through conformal prediction and conformal map. arXiv preprint arXiv:2505.16740 (2025)

work page arXiv 2025