Pith. sign in

REVIEW 3 major objections 8 minor 20 references

Epistemic ensemble disagreement separates nominal from perturbed gyro regimes more cleanly than learned aleatoric uncertainty in residual bias correction.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-31 10:43 UTC pith:WWXSOXXU

load-bearing objection Solid applied empirical study of residual-gyro CNN uncertainty and IG attribution; the FDIR claim outruns the evidence because OOD results are aggregated and star-tracker dropout sits in tension with near-zero ST attribution. the 3 major comments →

arxiv 2607.24608 v1 pith:WWXSOXXU submitted 2026-07-27 cs.LG

Attribution and Uncertainty Behavior of Learned Residual Gyro Correction for Gyro-Stellar Estimation

Mariela De Lucas \'Alvarez , Melvin Laux , Arthur de Freitas Precht , Maurice Martin , Edoardo Caroselli , Frank Kirchner , Alexander Fabisch This is my paper
classification cs.LG
keywords hybrid state estimationexplainable AIuncertainty quantificationgyro bias correctiondeep ensemblesintegrated gradientsaleatoric uncertaintyepistemic uncertainty
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Spacecraft attitude filters still leave residual gyroscope bias that a small external neural network can learn to cancel from gyro and star-tracker rates. This paper asks what that network’s uncertainties and explanations do when the sensors are hit by realistic, temporally structured degradation never seen in training. A 1-D CNN predicts both a mean rate correction and an input-dependent noise level; five independently trained copies supply an ensemble-disagreement (epistemic) signal. On simulator logs with AR noise, vibration, bias steps/drifts and star-tracker freezes of rising strength, aleatoric uncertainty grows but overlaps across regimes and is inconsistently calibrated, while ensemble variance stratifies cleanly and tightens the separation as the shift worsens. Gradient attributions show both the correction and the uncertainty are driven mainly by the gyroscope channels—especially the y-axis—and that this reliance pattern barely moves from clean to degraded data. The result matters because hybrid deep-filter pipelines can keep the classical estimator intact and still obtain a usable out-of-distribution monitor plus an account of which sensors are driving the elevated uncertainty.

Core claim

Under structured sensor perturbations unseen in training, epistemic uncertainty from deep-ensemble disagreement gives a clearer, more separable signal of distributional shift than the network’s heteroscedastic aleatoric uncertainty for residual gyro-rate correction, while Integrated Gradients attributions for both the mean correction and the log-variance stay dominated by gyroscope inputs (especially gyro-y) and remain largely stable from nominal to perturbed regimes.

What carries the argument

A residual 1-D CNN that jointly outputs mean correction μ and log-variance log σ² (Gaussian NLL plus weighted MSE), a five-member independent ensemble whose prediction variance is treated as epistemic uncertainty, and Integrated Gradients applied separately to μ and log σ² under AR(1), multi-tone vibration, bias-step/drift and hold-last-value dropout perturbations of three intensities.

Load-bearing premise

The synthetic structured perturbations and open-loop residual tests on simulator logs are representative enough of real spacecraft sensor degradation and closed-loop estimator behavior to support claims about monitoring and fault detection.

What would settle it

On real flight telemetry or closed-loop Gyro-Stellar Estimator runs with genuine sensor faults, if ensemble-variance distributions of nominal and degraded segments overlapped as heavily as the paper’s aleatoric distributions do, the claim that epistemic uncertainty is the better discriminator would be falsified.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Downstream monitors can treat rising ensemble disagreement as the primary flag for non-nominal sensor conditions and treat aleatoric rise as a secondary intensity cue.
  • Stable gyro-dominated attribution maps can be used to report which input channels are responsible when uncertainty climbs.
  • Residual correctors can remain external to the classical estimator without losing explainability or an OOD signal.
  • Aleatoric heads trained only on nominal data should not be assumed calibrated under structured degradation; regime-aware checks are required.
  • Axis-specific mid-window temporal peaks that persist under perturbation indicate the network reuses fixed temporal structure rather than redistributing attention.

Where Pith is reading between the lines

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

  • The same ensemble-disagreement monitor could be attached to other residual learners (star-tracker or actuator correctors) without touching the classical filter internals.
  • A controlled ablation that drops star-tracker channels at inference would test whether their near-zero attribution means they act only as a training-time reference rather than a true input dependence.
  • If the mid-window temporal peaks survive across different window lengths, the network may be learning a discrete-time bias estimator rather than a full dynamical model, which would simplify onboard assurance arguments.
  • Pairing the epistemic flag with a coverage check on standardized residuals could give operators a two-tier alert: shift detected versus confidence miscalibrated.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 8 minor

Summary. The manuscript presents an empirical study of uncertainty and attribution behavior in a hybrid state-estimation pipeline. A 1-D CNN with a heteroscedastic Gaussian head (mean residual correction μ and log-variance log σ², trained with an NLL+α·MSE loss) predicts residual gyroscope bias corrections from gyro and star-tracker rate windows; the correction is subtracted from gyro measurements upstream of a flight-representative Gyro-Stellar Estimator (GSE), which is left unmodified. Training uses nominal data from a proprietary high-fidelity mission simulator (100 logs, ~3.24 h each); evaluation is open-loop at the residual level under synthetic structured perturbations (AR(1) noise, multi-tone vibration, gyro bias step/drift, star-tracker hold-last-value dropout) at three hand-set intensities (OOD-L/M/H). Epistemic uncertainty is the variance of μ across an ensemble of M=5 independently trained models. Reported findings: (i) the CNN reduces mean residual magnitude by ~90% vs. raw and ~82% vs. a mean-bias-removal baseline (Table 2); (ii) aleatoric uncertainty rises with perturbation intensity but distributions overlap and calibration shifts from under-dispersion (ID) to over-dispersion (OOD); (iii) epistemic variance separates ID from OOD regimes more cleanly; (iv) Integrated Gradients attribution for both μ and log σ² is dominated by gyroscope channels (gyro-y especially), with negligible star-tracker contribution, and temporal/channel attribution structure remains stable across

Significance. If the results hold, the paper offers a useful, honest characterization of how heteroscedastic aleatoric and ensemble-epistemic uncertainty behave under temporally structured sensor degradation in a hybrid deep-filter architecture — a regime (correlated, physically motivated perturbations rather than i.i.d. noise) that prior UQ work on learned estimation components has largely skipped. The joint attribution of both the mean correction μ and the log-variance head log σ², aggregated per axis and per regime, is a genuinely uncommon analysis and the attribution-stability finding (uncertainty magnitude rises without temporal or channel redistribution) is an interpretable, falsifiable observation. The correction gains are concretely quantified (Table 2: ~90% mean residual reduction vs. raw, ~82% vs. naive baseline, with consistent tail improvements). The manuscript is also commendably careful about its own limits: it states the perturbations do not exhaust operational shifts and that results characterize controlled distribution shift rather than general OOD detection. It does not, however, ship code or data (the simulator is proprietary), and the downstream FDIR claim is motivational, so

major comments (3)
  1. [§3.3 (Perturbation Modeling), final paragraph] There is a direct contradiction about whether the perturbation data enters training. The second paragraph of §3.3 states 'The perturbation datasets are not used during training and are therefore out-of-distribution with respect to the training data,' but the final paragraph of §3.3 states 'we retrain the ensembles with these perturbations while maintaining the ground truth bias unchanged. This ensures that only the uncertainty awareness is tested to capture epistemic uncertainty through prediction disagreement.' If ensemble members are actually retrained on perturbed inputs, the OOD interpretation of Fig. 6 collapses, because epistemic separation rests on the perturbations being unseen. If (as I suspect) the intended meaning is that perturbed inputs are only passed through the already-trained ensembles at evaluation time, the sentence must be corrected, because as written it describes a
  2. [§4.4 / §4.5 / §3.3 (dropout), with §4.1 motivation] The manuscript's two result strands pull against each other at the point that matters most for the stated FDIR motivation. §3.3 defines the dropout perturbation as applying only to star trackers (hold-last-value); §4.5/Fig. 7 report that star tracker channels contribute 'negligibly' to both μ and log σ² across all regimes; and §4.1 motivates the whole study with a star tracker outage ('even a 10 min star tracker outage causes a ~1° attitude error'). If the network places near-zero functional weight on star tracker channels, a star-tracker-only corruption should produce little ensemble disagreement, i.e., little epistemic signal — so the clean separation in Fig. 6 may be driven entirely by the gyro-affecting perturbation families (AR noise, vibration, bias step/drift). Fig. 6 appears to pool all perturbation families into intensity levels OOD-L/M/H, so this cannot be checked from what is
  3. [§4.4 (Epistemic Uncertainty Behavior), Fig. 6] The central comparative claim — that epistemic uncertainty separates ID from OOD regimes with 'very small overlap' while aleatoric distributions overlap — is supported only visually (Fig. 4 vs. Fig. 6). Given that the ensemble has only M=5 members (so per-sample epistemic variance estimates are themselves noisy), a quantitative separation metric is needed: e.g., AUROC or a thresholded detection rate of a simple epistemic-variance classifier between ID and each OOD level, ideally with confidence intervals across evaluation logs. This is a modest addition but it is load-bearing for the paper's strongest claim and for any downstream threshold-based monitoring use.
minor comments (8)
  1. [§3.4 (Attribution Analysis)] The IG baseline is never specified (zeros, feature means, or other), and no completeness check (sum of attributions vs. output difference from baseline) is reported. Since the channel-dominance conclusions in Fig. 7 depend on attribution magnitudes, please state the baseline and ideally report the completeness error.
  2. [§4.3, z-score discussion] Wording: 'the residuals are spread out over a wide range, which means they are under-dispersed' is confusing — wide z-scores indicate that the predicted σ is too small, i.e., the predictive distribution is under-dispersed (equivalently the standardized residuals are over-dispersed). Please rephrase for precision.
  3. [§3.2, Eq. (2)] Eq. (2) defines epistemic uncertainty using only the spread of ensemble means, discarding the members' predicted variances. This is a reasonable choice given the stated goal of analyzing components separately, but please note explicitly that this is not the full predictive variance of a deep ensemble (Lakshminarayanan et al.) and that conclusions about 'epistemic' behavior are specific to this mean-variance definition.
  4. [Table 1] The dropout probability range is very narrow (6×10⁻⁴–8×10⁻⁴) compared to the wide ranges for the other parameters; please justify why probability is held nearly fixed while duration varies, or widen the sweep.
  5. [§4.1] Typo: 'Since, the onboard time is set at 8 Hz' — spurious comma.
  6. [Acknowledgments] Typos: 'European Aerospace Agency' should be 'European Space Agency'; 'lead by Airbus' should be 'led by Airbus'.
  7. [References [9], [5]] Duplicated URL string in the Kechris et al. entry; also Ref. [5] is a project webpage (ESA Nebula) rather than a publication — consider citing a peer-reviewed source for SHAP-based XAI in spacecraft GNC if one exists.
  8. [Abstract] Phrases such as 'epistemic uncertainty gives a clear signal that gets clearer as the distributional shift happens, showing that the models disagree more' are colloquial and slightly circular; suggest tightening to a quantitative statement once a separation metric is added.

Circularity Check

0 steps flagged

No significant circularity: empirical behavior study with standard UQ/XAI tools; results are measured, not forced by construction.

full rationale

This paper is an empirical analysis of a trained residual-correction CNN under held-out synthetic perturbations, not a first-principles derivation. The heteroscedastic head (Eq. 1 NLL+MSE), ensemble epistemic variance (Eq. 2), predictive entropy, and Integrated Gradients attributions are standard, externally established methods applied to new data; none redefine the reported ID/OOD separation or attribution patterns in terms of the quantities being claimed. Training is on nominal logs only; OOD regimes (AR(1), vibration, bias step/drift, dropout) are withheld and used for evaluation, so aleatoric growth, epistemic separation, calibration shifts, and gyro-dominated attributions are observational outcomes, not fitted inputs renamed as predictions. Citations (Kendall & Gal, Lakshminarayanan, Sundararajan IG, etc.) are external methodological support, not load-bearing self-citation uniqueness theorems. Internal tensions (e.g., negligible star-tracker attribution vs. dropout-only faults) are consistency/correctness concerns, not circular reductions. No step reduces a claimed prediction to its defining inputs by construction.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 0 invented entities

The central claims rest on standard probabilistic ML tools plus domain modeling choices for sensors and synthetic degradation. No new physical entities are postulated. Load-bearing free choices are the loss mix, windowing, ensemble size, BOHB-selected architecture, and hand-set perturbation schedules on proprietary simulator logs.

free parameters (5)
  • NLL+MSE mix weight α = not numerically reported
    Loss is Gaussian NLL plus weighted MSE; α is a training hyperparameter that shapes both μ accuracy and σ calibration.
  • Temporal window length and stride = 1.0 s window; stride 0.125–1.0 s explored
    Fixed at 1.0 s window after ablation; stride/overlap choices change sample dependence and attribution horizon.
  • Ensemble size M = M=5
    Epistemic signal is variance across M independently trained models; M directly defines the disagreement metric.
  • CNN architecture hyperparameters (layers, filters, kernel, pooling, dropout) = ranges in Table 1; single selected model
    Selected via BOHB on validation NLL; the reported behaviors are for one representative high-performing candidate.
  • Perturbation intensity schedules = e.g. ρ=0.05/0.10/0.20; drift 0.001/0.003/0.006 per s
    AR ρ, vibration scale, dropout duration/probability, bias step and drift rates define OOD-L/M/H and thus the claimed separation.
axioms (5)
  • domain assumption Observation noise is well modeled by a diagonal Gaussian heteroscedastic head with NLL training (Nix & Weigend / Kendall & Gal style).
    Calibration and entropy analyses assume Gaussian predictive intervals and z=(y-μ)/σ (§3.2).
  • domain assumption Variance across independently trained ensemble means is a valid operational proxy for epistemic uncertainty / OOD disagreement.
    Eq. (2) and §4.4 treat ensemble variance as the epistemic signal without Bayesian posterior guarantees.
  • standard math Integrated Gradients attributions (with standard sensitivity/implementation-invariance axioms) indicate model reliance for μ and log σ² on time-series inputs.
    §3.4 and related work invoke IG as the explainability backbone; baseline choice details are thin.
  • ad hoc to paper Synthetic AR(1), vibration, bias step/drift, and hold-last-value dropout are realistic proxies for spacecraft sensor degradation while preserving underlying dynamics.
    §3.3 explicitly builds OOD regimes this way and caveats that they do not exhaust operational shifts.
  • domain assumption Open-loop residual-level evaluation isolates learned-module behavior without needing closed-loop GSE interaction effects.
    §3.1 states open-loop residual evaluation to avoid confounding; conclusions about monitoring still gesture at downstream estimators.

pith-pipeline@v1.2.0-grok45-kimik3 · 17503 in / 3603 out tokens · 71777 ms · 2026-07-31T10:43:51.194823+00:00 · methodology

0 comments
read the original abstract

This work investigates uncertainty decomposition and explainability in a deep learning-based framework for gyroscope bias correction. A 1-D Convolutional Neural Network is trained to predict residual angular rate corrections from multi-sensor inputs, including gyroscope and star tracker measurements. The bias corrections are sent to a flight-representative Gyro-Stellar Estimator. The network produces both mean corrections and input-dependent (heteroscedastic) aleatoric uncertainty, while epistemic uncertainty is estimated via an ensemble of independently trained models. The proposed approach is trained under nominal conditions and evaluated in both nominal and structured perturbations that include additive and temporally correlated noise. Gradient-based attribution methods are applied to both the correction and uncertainty outputs, enabling a decomposition of the evidence that drives state updates and uncertainty estimates. By aggregating attribution patterns across rotational axes and regimes, we reveal axis-specific behaviors and characterize how structured perturbations influence the collaboration between aleatoric and epistemic uncertainty. Uncertainty analysis shows that aleatoric uncertainty increases with perturbation intensity, but the distributions overlap and the calibration is not consistent across regimes. On the other hand, epistemic uncertainty gives a clear signal that gets clearer as the distributional shift happens, showing that the models disagree more. These results show that aleatoric and epistemic uncertainty work well together and that epistemic uncertainty is better at distinguishing between nominal and perturbed operating conditions. The results provide insight into the behavior of hybrid learning-based state estimation components and motivate the use of uncertainty for downstream monitoring and fault detection.

Figures

Figures reproduced from arXiv: 2607.24608 by Alexander Fabisch, Arthur de Freitas Precht, Edoardo Caroselli, Frank Kirchner, Mariela De Lucas \'Alvarez, Maurice Martin, Melvin Laux.

Figure 1
Figure 1. Figure 1: Overview of the analyzed hybrid estimation pipeline. A 1D-CNN processes [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Gyro-only drift analysis across 100 logs. a) Rotation error grows mono [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Residual distribution per-axis. NN-based filter correction reduces the [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Aleatoric uncertainty across regimes. a) Predicted entropy increases with [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Standardized residuals (z-scores) distributions across regimes. a) Residual distributions become progressively narrower as perturbation intensity increases, indicating increasingly conservative uncertainty estimates under OOD condi￾tions. b) Sample-wise z-scores show regime-dependent shifts in residual spread and concentration. When conditions are normal, the residuals are spread out over a wide range, whi… view at source ↗
Figure 6
Figure 6. Figure 6: log10 epistemic uncertainty by regime. OOD distributions progressively shift toward higher uncertainty values, indicating increased ensemble disagree￾ment under stronger perturbations. 4.4 Epistemic Uncertainty Behavior Epistemic uncertainty is estimated using an ensemble of independently trained models. we calculate it as the variance of the predicted corrections across en￾semble members, capturing model … view at source ↗
Figure 7
Figure 7. Figure 7: Channel attribution is dominated by gyro inputs in all regimes, partic [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Temporal attribution is mainly concentrated in central timesteps, not [PITH_FULL_IMAGE:figures/full_fig_p017_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Representative single-sample IG attribution maps for the predictive mean [PITH_FULL_IMAGE:figures/full_fig_p018_9.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

20 extracted references · 2 canonical work pages

  1. [1]

    IEEE Transactions on Instrumen- tation and Measurement69(1), 24–34 (2020).https://doi.org/10.1109/TIM

    Al-Sharman, M.K., Zweiri, Y., Jaradat, M.A.K., Al-Husari, R., Gan, D., Senevi- ratne, L.D.: Deep-learning-based neural network training for state estimation en- hancement: Application to attitude estimation. IEEE Transactions on Instrumen- tation and Measurement69(1), 24–34 (2020).https://doi.org/10.1109/TIM. 2019.2895495

    arXiv 2020

  2. [2]

    PLoS ONE10(7), e0130140 (2015).https://doi.org/10

    Bach, S., Binder, A., Montavon, G., Klauschen, F., Müller, K.R., Samek, W.: On pixel-wise explanations for non-linear classifier decisions by layer-wise rel- evance propagation. PLoS ONE10(7), e0130140 (2015).https://doi.org/10. 1371/journal.pone.0130140

    2015

  3. [3]

    Springer-Verlag New York, Inc., Secaucus, NJ, USA (2006)

    Bishop, C.M.: Pattern Recognition and Machine Learning (Information Science and Statistics). Springer-Verlag New York, Inc., Secaucus, NJ, USA (2006)

    2006

  4. [4]

    Journal of Time Series Analysis37(5), 709–711 (sep 2016)

    Box, G.E., Jenkins, G.M., Reinsel, G.C., Ljung, G.M.: Time Series Analysis: Fore- casting and Control. Journal of Time Series Analysis37(5), 709–711 (sep 2016). https://doi.org/10.1111/jtsa.12194

    doi:10.1111/jtsa.12194 2016

  5. [5]

    City University of London: Explainable secure deep learning software for spacecraft GNC systems. Tech. rep. (2024),https://nebula.esa.int/content/ explainable-secure-deep-learning-software-spacecraft-gnc-systems

    2024

  6. [6]

    Cover, T.M., Thomas, J.A.: Elements of Information Theory (2005)

    2005

  7. [7]

    Falkner, S., Klein, A., Hutter, F.: BOHB: Robust and Efficient Hyperparameter Optimization at Scale. Tech. rep. (2018)

    2018

  8. [8]

    Proceedings of Machine Learning Research267, 26877– 26895 (jun 2025),https://arxiv.org/pdf/2506.05035

    Jang, H., Kim, C., Yang, E.: TIMING: Temporality-Aware Integrated Gradients for Time Series Explanation. Proceedings of Machine Learning Research267, 26877– 26895 (jun 2025),https://arxiv.org/pdf/2506.05035

    Pith/arXiv arXiv 2025

  9. [9]

    Kechris, C., Dan, J., Atienza, D.: Time series saliency maps: explaining models across multiple domains (may 2025),https://arxiv.org/pdf/2505.13100https: //arxiv.org/pdf/2505.13100v1

    Pith/arXiv arXiv 2025

  10. [10]

    Kendall, A., Gal, Y.: What Uncertainties Do We Need in Bayesian Deep Learning for Computer Vision? Advances in Neural Information Processing Systems2017- December, 5575–5585 (mar 2017),https://arxiv.org/pdf/1703.04977

    Pith/arXiv arXiv 2017

  11. [11]

    In: 35th International Conference on Machine Learning, ICML 2018

    Kuleshov, V., Fenner, N., Ermon, S.: Accurate uncertainties for deep learning us- ing calibrated regression. In: 35th International Conference on Machine Learning, ICML 2018. vol. 6, pp. 4369–4377 (2018)

    2018

  12. [12]

    Advances in Neural Information Processing Systems pp

    Lakshminarayanan, B., Pritzel, A., Blundell, C.: Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles. Advances in Neural Information Processing Systems pp. 6403–6414 (dec 2016),https://arxiv.org/pdf/1612. 01474

    2016

  13. [13]

    Applied Mechanics Reviews41(5), 187–187 (may 1988).https://doi.org/10.1115/1.3151890

    Lalanne, C.: Random Vibration. Applied Mechanics Reviews41(5), 187–187 (may 1988).https://doi.org/10.1115/1.3151890

    work page doi:10.1115/1.3151890 1988

  14. [14]

    In: Proceedings of the 31st International Conference on Neural Information Processing Systems

    Lundberg, S.M., Lee, S.I.: A unified approach to interpreting model predictions. In: Proceedings of the 31st International Conference on Neural Information Processing Systems. pp. 4768–4777. NIPS’17, Curran Associates Inc., Red Hook, NY, USA (2017)

    2017

  15. [15]

    In: International Conference on Agents and Artificial Intelligence

    Mercier, D., Bhatt, J., Dengel, A., Ahmed, S.: Time to Focus: A Comprehensive Benchmark using Time Series Attribution Methods. In: International Conference on Agents and Artificial Intelligence. vol. 2, pp. 562–573. Science and Technology Publications, Lda (feb 2022).https://doi.org/10.5220/0010904400003116

    work page doi:10.5220/0010904400003116 2022

  16. [16]

    In: Proceedings of 1994 IEEE International Conference on Neural Attribution and Uncertainty in Gyro Bias Correction 21 Networks (ICNN’94)

    Nix, D., Weigend, A.: Estimating the mean and variance of the target probability distribution. In: Proceedings of 1994 IEEE International Conference on Neural Attribution and Uncertainty in Gyro Bias Correction 21 Networks (ICNN’94). vol. 1, pp. 55–60 vol.1. IEEE (1994).https://doi.org/10. 1109/ICNN.1994.374138

    arXiv 1994

  17. [17]

    IEEE Transactions on Signal Processing70, 1532–1547 (2022).https: //doi.org/10.1109/TSP.2022.3158588

    Revach, G., Shlezinger, N., Ni, X., Escoriza, A.L., Van Sloun, R.J., Eldar, Y.C.: KalmanNet: Neural Network Aided Kalman Filtering for Partially Known Dy- namics. IEEE Transactions on Signal Processing70, 1532–1547 (2022).https: //doi.org/10.1109/TSP.2022.3158588

    arXiv 2022

  18. [18]

    International Journal of Computer Vision128(2), 336–359 (2020).https://doi

    Selvaraju, R.R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., Batra, D.: Grad- CAM: Visual explanations from deep networks via gradient-based localization. International Journal of Computer Vision128(2), 336–359 (2020).https://doi. org/10.1007/s11263-019-01228-7

    doi:10.1007/s11263-019-01228-7 2020

  19. [19]

    In: 34th International Conference on Machine Learning, ICML 2017

    Sundararajan, M., Taly, A., Yan, Q.: Axiomatic attribution for deep networks. In: 34th International Conference on Machine Learning, ICML 2017. vol. 7, pp. 5109–5118. International Machine Learning Society (IMLS) (jun 2017).https: //doi.org/10.48550/arxiv.1703.01365

    Pith/arXiv arXiv doi:10.48550/arxiv.1703.01365 2017

  20. [20]

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

    Wang, H., Joshi, D., Wang, S., Ji, Q.: Gradient-based Uncertainty Attribution for Explainable Bayesian Deep Learning. In: Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition. vol. 2023-June, pp. 12044–12053 (2023).https://doi.org/10.1109/CVPR52729.2023.01159

    arXiv 2023