pith. sign in

arxiv: 2605.21309 · v1 · pith:WOHLHC6Dnew · submitted 2026-05-20 · 💻 cs.CV · cs.RO

Hyper-V2X: Hypernetworks for Estimating Epistemic and Aleatoric Uncertainty in Cooperative Bird's-Eye-View Semantic Segmentation

Abhishek Dinkar Jagtap , Sanath Tiptur Sadashivaiah , Andreas Festag This is my paper

Pith reviewed 2026-05-21 04:56 UTC · model grok-4.3

classification 💻 cs.CV cs.RO
keywords cooperative perceptionV2X communicationuncertainty estimationepistemic uncertaintyaleatoric uncertaintybird's-eye-view segmentationhypernetworksautonomous driving
0
0 comments X

The pith

Hyper-V2X conditions a Bayesian hypernetwork on fused multi-agent features to estimate both epistemic and aleatoric uncertainty in cooperative BEV semantic segmentation.

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

The paper introduces Hyper-V2X as a way to add uncertainty quantification to V2X cooperative perception for autonomous driving. It uses a partial weight generation scheme inside a Bayesian hypernetwork, conditioned by a V2X context embedding on fused data from multiple vehicles, to produce stochastic weights for bird's-eye-view segmentation. This produces separate estimates of epistemic uncertainty, which reflects model ignorance, and aleatoric uncertainty, which reflects data noise. A sympathetic reader would care because reliable uncertainty scores could flag unreliable predictions in shared environmental models, potentially supporting safer decisions without large extra computation.

Core claim

Hyper-V2X proposes a partial weight generation scheme and V2X context embedding module that conditions a Bayesian hypernetwork on fused multi-agent features to generate weight distributions for stochastic Bird's-Eye-View segmentation, enabling efficient estimation of both epistemic and aleatoric uncertainties in V2X-based perception while remaining architecture-agnostic.

What carries the argument

The Bayesian hypernetwork with partial weight generation conditioned via V2X context embedding, which takes fused multi-agent features and produces distributions over segmentation model weights to quantify uncertainties.

If this is right

  • Hyper-V2X improves overall perception reliability in cooperative V2X settings.
  • It delivers accurate, well-calibrated estimates of both epistemic and aleatoric uncertainty.
  • The approach adds little computation overhead relative to deterministic BEV models.
  • It integrates with existing cooperative backbones such as CoBEVT without requiring architecture changes.

Where Pith is reading between the lines

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

  • The same hypernetwork conditioning could extend to other cooperative tasks such as object detection or trajectory prediction.
  • Downstream planners could use the separated uncertainty types to adjust risk thresholds differently for model gaps versus sensor noise.
  • Testing the method on datasets with more variable numbers of communicating agents would check how well the context embedding scales.

Load-bearing premise

The fused multi-agent features after V2X context embedding contain enough information for the Bayesian hypernetwork's partial weight generation to separate epistemic from aleatoric uncertainty without systematic bias.

What would settle it

If experiments on the OPV2V benchmark show that uncertainty estimates are poorly calibrated or that perception reliability does not improve over deterministic baselines, the central claim would be falsified.

Figures

Figures reproduced from arXiv: 2605.21309 by Abhishek Dinkar Jagtap, Andreas Festag, Sanath Tiptur Sadashivaiah.

Figure 1
Figure 1. Figure 1: Overview of the proposed Hyper-V2X framework for uncertainty estimation in V2X-based cooperative perception. then given by the expectation of the loss over different sampled weights and is characterized by: θ ∼ qϕ(θ | c), c ∈ C, θ ∈ Θ, (5) LBHN(ϕ) = Eθ∼qϕ(θ|c) [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Qualitative results on OPV2V benchmark. Ground truth, predicted BEV segmentation, and corresponding epistemic and aleatoric uncertainty maps for representative scenes. Bm M m=1 and calculating the weighted average gap between accuracy and confidence: ECE = X M m=1 |Bm| N , [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Uncertainty estimation under varying compression rates. As CPR increases (0→64), segmentation quality degrades (red circles). Our method produces uncertainty maps that effectively capture this degradation, with progressively higher epistemic and aleatoric uncertainty in vulnerable regions, demonstrating reliable uncertainty estimation under communication constraints. at compression rate 0 (highlighted in r… view at source ↗
read the original abstract

Cooperative perception enabled by Vehicle-to-Everything (V2X) communication enhances autonomous driving safety by creating a unified environmental representation through shared sensory data. While recent works have advanced multi-agent fusion for improved perception, uncertainty quantification in such cooperative frameworks remains largely unexplored. This paper introduces Hyper-V2X, a hypernetwork-based framework for estimating both epistemic and aleatoric uncertainties in V2X-based perception. Specifically, we propose a partial weight generation scheme and V2X context embedding module that conditions a Bayesian hypernetwork on fused multi-agent features to generate weight distributions for stochastic Bird's-Eye-View (BEV) segmentation. Unlike existing deterministic BEV models, Hyper-V2X enables efficient uncertainty estimation with little computation overhead. Our approach is architecture-agnostic, and can be seamlessly integrating with modern cooperative backbones such as CoBEVT. Experiments on the OPV2V benchmark demonstrate that Hyper-V2X provides accurate, well-calibrated uncertainty estimates and improves overall perception reliability. Our code and benchmark are publicly available under an open-source license: https://github.com/abhishekjagtap1/Hyper-V2X

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.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces Hyper-V2X, a hypernetwork-based framework for estimating epistemic and aleatoric uncertainties in cooperative Bird's-Eye-View semantic segmentation for V2X perception. It proposes a partial weight generation scheme together with a V2X context embedding module that conditions a Bayesian hypernetwork on fused multi-agent features to produce stochastic BEV segmentation outputs. The method is presented as architecture-agnostic and compatible with backbones such as CoBEVT, with experiments on the OPV2V benchmark claimed to yield accurate, well-calibrated uncertainty estimates that improve overall perception reliability at negligible computational cost.

Significance. If the central claims are substantiated, the work would address an important and largely unexplored area of uncertainty quantification in multi-agent cooperative perception, directly relevant to safety-critical autonomous driving. The architecture-agnostic design and the public release of code and benchmarks constitute clear strengths that support reproducibility and potential adoption by the community.

major comments (2)
  1. [§3.2] §3.2 (Partial Weight Generation): The claim that conditioning the Bayesian hypernetwork on V2X context-embedded fused features cleanly separates epistemic from aleatoric uncertainty without systematic bias or correlation is load-bearing for the assertion of accurate, well-calibrated estimates without post-hoc calibration. The manuscript provides no formal argument, ablation study, or diagnostic (e.g., correlation between the two uncertainty maps) demonstrating that the partial generation scheme achieves this orthogonality in the presence of V2X fusion.
  2. [§5] §5 (Experiments): The abstract states that Hyper-V2X supplies accurate and well-calibrated uncertainty estimates together with reliability gains, yet the results must include concrete quantitative support—such as reliability diagrams, expected calibration error values, per-uncertainty-type metrics, and comparisons to deterministic CoBEVT and other uncertainty baselines—complete with error bars across multiple runs to substantiate the central claims.
minor comments (2)
  1. [Abstract] Abstract: the phrase 'can be seamlessly integrating with' is grammatically incorrect and should read 'can be seamlessly integrated with'.
  2. Notation: the distinction between the hypernetwork parameters and the generated weight distributions should be made explicit with consistent symbols throughout the method section to avoid reader confusion.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and for recognizing the relevance of uncertainty quantification in cooperative V2X perception. We address the major comments point by point below and will incorporate the suggested improvements in the revised manuscript.

read point-by-point responses

  1. Referee: [§3.2] §3.2 (Partial Weight Generation): The claim that conditioning the Bayesian hypernetwork on V2X context-embedded fused features cleanly separates epistemic from aleatoric uncertainty without systematic bias or correlation is load-bearing for the assertion of accurate, well-calibrated estimates without post-hoc calibration. The manuscript provides no formal argument, ablation study, or diagnostic (e.g., correlation between the two uncertainty maps) demonstrating that the partial generation scheme achieves this orthogonality in the presence of V2X fusion.

    Authors: We acknowledge that the current manuscript does not include an explicit formal argument or diagnostic analysis (such as correlation coefficients between the two uncertainty maps) to demonstrate orthogonality. The partial weight generation scheme is designed so that the Bayesian hypernetwork models epistemic uncertainty via stochastic weight sampling while the V2X context embedding modulates the generated weights to reflect input-dependent variations associated with aleatoric uncertainty. To strengthen this claim, we will add an ablation study with correlation diagnostics and additional visualizations of the two uncertainty types in the revised version. revision: yes

  2. Referee: [§5] §5 (Experiments): The abstract states that Hyper-V2X supplies accurate and well-calibrated uncertainty estimates together with reliability gains, yet the results must include concrete quantitative support—such as reliability diagrams, expected calibration error values, per-uncertainty-type metrics, and comparisons to deterministic CoBEVT and other uncertainty baselines—complete with error bars across multiple runs to substantiate the central claims.

    Authors: We agree that the experimental section would benefit from more comprehensive quantitative validation to support the claims of calibration and reliability gains. While the OPV2V results demonstrate improved perception reliability, we will expand the experiments to include reliability diagrams, Expected Calibration Error (ECE) values computed separately for epistemic and aleatoric uncertainties, per-uncertainty-type metrics, direct comparisons against deterministic CoBEVT and other uncertainty baselines, and results reported as means with standard deviations across multiple runs with error bars. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation relies on external benchmark and independent architectural choices

full rationale

The paper proposes a hypernetwork framework with partial weight generation and V2X context embedding to estimate epistemic and aleatoric uncertainty in cooperative BEV segmentation. No equations, derivations, or self-citations in the abstract or described approach reduce the claimed uncertainty estimates to quantities defined by fitted parameters or prior self-referential results within the same paper. The method is presented as architecture-agnostic, integrable with external backbones such as CoBEVT, and evaluated on the independent OPV2V benchmark, keeping the central claims self-contained against external validation rather than internally forced by construction.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

Review performed on abstract only; therefore the ledger reflects typical deep-learning assumptions rather than paper-specific derivations. The central claim rests on the effectiveness of feature fusion and hypernetwork conditioning, which are not derived from first principles.

free parameters (1)
  • Hypernetwork and embedding parameters
    Weights of the hypernetwork and V2X context embedding module are learned from OPV2V training data; exact count and initialization not stated in abstract.
axioms (1)
  • domain assumption Fused multi-agent features after context embedding contain sufficient information to condition weight distributions for reliable epistemic/aleatoric separation.
    Invoked by the description of the V2X context embedding module and partial weight generation scheme.

pith-pipeline@v0.9.0 · 5759 in / 1533 out tokens · 48626 ms · 2026-05-21T04:56:05.241363+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages · 1 internal anchor

  1. [1]

    Systematic literature review on vehicular collaborative perception – A computer vision perspective,

    L. Wanet al., “Systematic literature review on vehicular collaborative perception – A computer vision perspective,”Accepted for IEEE Trans. Intell. Transp. Syst., 2025, doi: 10.48550/arXiv.2504.04631

    work page doi:10.48550/arxiv.2504.04631 2025

  2. [2]

    Cooper: Cooperative perception for connected autonomous vehicles based on 3D point clouds,

    Q. Chen, S. Tang, Q. Yang, and S. Fu, “Cooper: Cooperative perception for connected autonomous vehicles based on 3D point clouds,” inIEEE Intern. Conf. on Distr. Comp. Syst. (ICDCS), 2019, doi: 10.1109/ICDCS.2019.00058

    work page doi:10.1109/icdcs.2019.00058 2019

  3. [3]

    Where2comm: Communication-efficient collaborative perception via spatial confidence maps,

    Y . Hu, S. Fang, Z. Lei, Y . Zhong, and S. Chen, “Where2comm: Communication-efficient collaborative perception via spatial confidence maps,” inAdv. Neural Inform. Process. Syst. (NeurIPS), 2022. [Online]. Available: https://proceedings.neurips.cc/paper files/paper/ 2022/file/1f5c5cd01b864d53cc5fa0a3472e152e-Paper-Conference.pdf

    work page 2022

  4. [4]

    How2comm: Communication-efficient and collaboration-pragmatic multi-agent perception,

    D. Yanget al., “How2comm: Communication-efficient and collaboration-pragmatic multi-agent perception,” inAdv. Neural Inform. Process. Syst. (NeurIPS), 2023. [On- line]. Available: https://proceedings.neurips.cc/paper files/paper/2023/ file/4f31327e046913c7238d5b671f5d820e-Paper-Conference.pdf

    work page 2023

  5. [5]

    Improving Infrastructure and Community Resilience with Shared Autonomous Electric Vehicles (SAEV-R),

    Q. Delooz, A. Festag, A. Vinel, and S. C. Lobo, “Simulation- based performance optimization of V2X collective perception by adaptive object filtering,” inIEEE IV Symposium, 2023, doi: 10.1109/IV55152.2023.10186788

    work page doi:10.1109/iv55152.2023.10186788 2023

  6. [6]

    First Mile: An open innovation lab for infrastructure- assisted cooperative intelligent transportation systems,

    R. Songet al., “First Mile: An open innovation lab for infrastructure- assisted cooperative intelligent transportation systems,” inIEEE IV Symposium, 2024, doi: 10.1109/IV55156.2024.10588500

    work page doi:10.1109/iv55156.2024.10588500 2024

  7. [7]

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

    Y . Huet al., “Collaboration helps camera overtake LiDAR in 3D detection,” inIEEE/CVF Conf. Comput. Vis. Pattern Recog. (CVPR), 2023, doi: 10.1109/CVPR52729.2023.00892

    work page doi:10.1109/cvpr52729.2023.00892 2023

  8. [8]

    V2X-ViT: Vehicle-to-everything cooperative perception with vision transformer,

    R. Xuet al., “V2X-ViT: Vehicle-to-everything cooperative perception with vision transformer,” inEur. Conf. Comput. Vis. (ECCV). Springer, 2022, doi: 10.1007/978-3-031-19842-7 7

    work page doi:10.1007/978-3-031-19842-7 2022

  9. [9]

    CoBEVT: Cooperative bird’s eye view semantic segmentation with sparse transformers,

    ——, “CoBEVT: Cooperative bird’s eye view semantic segmentation with sparse transformers,” inConference on Robot Learning (CoRL),

    work page

  10. [10]

    Available: https://proceedings.mlr.press/v205/xu23a/ xu23a.pdf

    [Online]. Available: https://proceedings.mlr.press/v205/xu23a/ xu23a.pdf

    work page

  11. [11]

    Freeman, Frédo Durand, Eli Shechtman, and Xun Huang

    J. Fuet al., “Generative map priors for collaborative BEV semantic segmentation,” inIEEE/CVF Conf. Comput. Vis. Pattern Recog. (CVPR), 2025, doi: 10.1109/CVPR52734.2025.01113

    work page doi:10.1109/cvpr52734.2025.01113 2025

  12. [12]

    categories

    R. Songet al., “Collaborative semantic occupancy prediction with hybrid feature fusion in connected automated vehicles,” in IEEE/CVF Conf. Comput. Vis. Pattern Recog. (CVPR), 2024, doi: 10.1109/CVPR52733.2024.01704

    work page doi:10.1109/cvpr52733.2024.01704 2024

  13. [13]

    Toward collaborative autonomous driving: Simulation platform and end-to-end system,

    G. Liuet al., “Toward collaborative autonomous driving: Simulation platform and end-to-end system,”IEEE Trans. Pattern Anal. Mach. Intell., vol. 47, no. 8, 2025, doi: 10.1109/TPAMI.2025.3560327

    work page doi:10.1109/tpami.2025.3560327 2025

  14. [14]

    V2X-Gaussians: Gaussian splatting for multi-agent cooperative dy- namic scene reconstruction,

    A. D. Jagtap, R. Song, S. T. Sadashivaiah, and A. Festag, “V2X-Gaussians: Gaussian splatting for multi-agent cooperative dy- namic scene reconstruction,” in2025 IEEE IV Symposium, 2025, doi: 10.1109/IV64158.2025.11097436

    work page doi:10.1109/iv64158.2025.11097436 2025

  15. [15]

    Bayesian hypernetworks,

    D. Kruegeret al., “Bayesian hypernetworks,” inNIPS Workshop Bayesian Deep Learning, 2017. [Online]. Available: https://bayesiandeeplearning.org/2017/papers/34.pdf

    work page 2017

  16. [16]

    Dropout as a bayesian approximation: Representing model uncertainty in deep learning,

    Y . Gal and Z. Ghahramani, “Dropout as a bayesian approximation: Representing model uncertainty in deep learning,” inInter. Conf. on Mach. Learn. (ICML), 2016. [Online]. Available: https://proceedings. mlr.press/v48/gal16.html

    work page 2016

  17. [17]

    Weight uncertainty in neural networks,

    C. Blundell, J. Cornebise, K. Kavukcuoglu, and D. Wierstra, “Weight uncertainty in neural networks,” inInter. Conf. on Mach. Learn. (ICML), 2015. [Online]. Available: http://proceedings.mlr.press/v37/ blundell15.html

    work page 2015

  18. [18]

    Simple and scalable predictive uncertainty estimation using deep ensembles,

    B. Lakshminarayanan, A. Pritzel, and C. Blundell, “Simple and scalable predictive uncertainty estimation using deep ensembles,” inAdv. Neural Inform. Process. Syst. (NeurIPS), 2017. [Online]. Available: https://proceedings.neurips.cc/paper files/paper/ 2017/file/9ef2ed4b7fd2c810847ffa5fa85bce38-Paper.pdf

    work page 2017

  19. [19]

    ImmFusion: Robust mmWave-RGB Fusion for 3D Human Body Reconstruction in All Weather Conditions

    S. Suet al., “Uncertainty quantification of collaborative detection for self-driving,” inIEEE Inter. Conf. on Rob. and Autom (ICRA), 2023, doi: 10.1109/ICRA48891.2023.10160367

    work page doi:10.1109/icra48891.2023.10160367 2023

  20. [20]

    Uncer- tainty quantification for collaborative object detection under adversarial attacks,

    H. Huang, C. Chen, J.-P. Monteuuis, J. Petit, and F. Miao, “Uncer- tainty quantification for collaborative object detection under adversarial attacks,”arXiv:2502.02537, 2025

    work page arXiv 2025

  21. [21]

    V2VNet: Vehicle-to-vehicle communication for joint perception and prediction,

    T.-H. Wanget al., “V2VNet: Vehicle-to-vehicle communication for joint perception and prediction,” inEur. Conf. Comput. Vis. (ECCV). Springer, Aug. 2020, DOI: 10.1007/978-3-030-58536-5 36

    work page doi:10.1007/978-3-030-58536-5 2020

  22. [22]

    Multiplicative normalizing flows for variational Bayesian neural networks,

    C. Louizos and M. Welling, “Multiplicative normalizing flows for variational Bayesian neural networks,” inInter. Conf. on Mach. Learn. (ICML). PMLR, 2017. [Online]. Available: https: //proceedings.mlr.press/v70/louizos17a/louizos17a.pdf

    work page 2017

  23. [23]

    Estimating epistemic and aleatoric uncertainty with a single model,

    M. Chan, M. Molina, and C. Metzler, “Estimating epistemic and aleatoric uncertainty with a single model,” inAdv. Neural Inform. Process. Syst. (NeurIPS), 2024. [Online]. Available: https://proceedings.neurips.cc/paper files/paper/ 2024/file/c693c3ff83259aebcd55a41ab19a5d84-Paper-Conference.pdf

    work page 2024

  24. [24]

    Collaborative multi-object tracking with conformal un- certainty propagation,

    S. Suet al., “Collaborative multi-object tracking with conformal un- certainty propagation,”IEEE Robot. Autom. L., vol. 9, no. 4, 2024, doi: 10.1109/LRA.2024.3364450

    work page doi:10.1109/lra.2024.3364450 2024

  25. [25]

    BEVFormer: Learning bird’s-eye-view representa- tion from LiDAR-camera via spatiotemporal transformers,

    Z. Liet al., “BEVFormer: Learning bird’s-eye-view representa- tion from LiDAR-camera via spatiotemporal transformers,”IEEE Trans. Pattern Anal. Mach. Intell., vol. 47, no. 03, 2025, doi: 10.1109/TPAMI.2024.3515454

    work page doi:10.1109/tpami.2024.3515454 2025

  26. [26]

    Motion planning for autonomous driving: The state of the art and future perspectives,

    C. Changet al., “BEV-V2X: Cooperative birds-eye-view fusion and grid occupancy prediction via v2x-based data sharing,”IEEE Transactions on Intelligent Vehicles, vol. 8, no. 11, 2023, doi:10.1109/TIV .2023.3293954

    work page doi:10.1109/tiv 2023

  27. [27]

    Evaluating uncertainty quantification for bird’s eye view semantic segmentation,

    B. Yanget al., “Evaluating uncertainty quantification for bird’s eye view semantic segmentation,” inWorkshop on Uncertainty Reasoning and Quantification in Decision Making, 2023. [Online]. Available: https://charliezhaoyinpeng.github.io/UDM-KDD23/ap/

    work page 2023

  28. [28]

    HyperNetworks

    D. Ha, A. M. Dai, and Q. V . Le, “Hypernetworks,” inInt. Conf. Learn. Represent. (ICLR), 2017, doi: 10.48550/arXiv.1609.09106

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1609.09106 2017

  29. [29]

    Partial hypernetworks for continual learning,

    H. Hemati, V . Lomonaco, D. Bacciu, and D. Borth, “Partial hypernetworks for continual learning,” inConference on Lifelong Learning Agents (CoLLAs), 2023. [Online]. Available: https://proceedings.mlr.press/v232/hemati23a.html

    work page 2023

  30. [30]

    F-cooper: Feature based cooperative perception for autonomous vehicle edge computing system using 3D point clouds,

    Q. Chenet al., “F-cooper: Feature based cooperative perception for autonomous vehicle edge computing system using 3D point clouds,” inACM/IEEE Symposium on Edge Computing, 2019, doi: 10.1145/3318216.3363300

    work page doi:10.1145/3318216.3363300 2019

  31. [31]

    Varadarajan, A

    R. Xuet al., “OPV2V: An open benchmark dataset and fu- sion pipeline for perception with vehicle-to-vehicle communica- tion,” inIEEE Inter. Conf. on Rob. and Autom (ICRA), 2022, doi: 10.1109/ICRA46639.2022.9812038

    work page doi:10.1109/icra46639.2022.9812038 2022

  32. [32]

    Learning distilled collaboration graph for multi- agent perception,

    Y . Liet al., “Learning distilled collaboration graph for multi- agent perception,” inAdv. Neural Inform. Process. Syst. (NeurIPS),

    work page

  33. [33]

    Available: https://proceedings.neurips.cc/paper files/ paper/2021/hash/f702defbc67edb455949f46babab0c18-Abstract.html

    [Online]. Available: https://proceedings.neurips.cc/paper files/ paper/2021/hash/f702defbc67edb455949f46babab0c18-Abstract.html

    work page 2021

  34. [34]

    Extracting uncertainty estimates from mixtures of experts for semantic segmentation,

    S. Pavlitska, B. Keskin, A. Faßbender, C. Hubschneider, and J. M. Z¨ollner, “Extracting uncertainty estimates from mixtures of experts for semantic segmentation,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, October 2025, pp. 311–320

    work page 2025

  35. [35]

    A stochastic approximation method

    S. Kullback and R. A. Leibler, “On information and sufficiency,”The Annals of Mathematical Statistics, vol. 22, no. 1, pp. 79–86, 1951, doi: 10.1214/aoms/1177729694

    work page doi:10.1214/aoms/1177729694 1951