Pith. sign in

REVIEW 3 major objections 5 minor 53 references

Plug-and-Play Reweighting for Resilient Collaborative Decision-Making in Connected Autonomous Driving

T0 review · 3 major / 5 minor · reviewed 2026-07-14 · grok-4.5

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

Pith's one-line read A plug-and-play reweighting step that down-weights neighborhood points far from the local median makes attention-based collaborative braking decisions resilient to noise and attacks without retraining.

desk verdict Solid plug-and-play robust-attention trick for collab LiDAR braking in sim; real gains, thin baselines, and the median assumption is the load-bearing soft spot. read the letter →

arxiv 2607.10037 v1 pith:SSLNULBQ submitted 2026-07-10 cs.RO

Jiewen Liu , Rui Liu , Matthew Lee , Ming C. Lin , Xiaorui Liu , Peng Gao This is my paper
classification cs.RO
keywords collaborativedecision-makingconnectedautonomousdrivingplug-and-playreweightingattention-basedfusionLiDARresilienceadversarialattacksmulti-robotsystems
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Connected vehicles need to fuse each other's LiDAR observations to decide whether to brake in occluded, accident-prone scenes, yet noise or malicious edits can make those shared points misleading. This paper claims that an attention encoder-decoder for collaborative decision-making can be made resilient by a training-free reweighting step: for each point, measure how far its neighbors' value embeddings sit from the group median, then shrink the attention weights of those that deviate strongly. The same module works both when a vehicle embeds its own noisy cloud and when the ego vehicle fuses collaborators' messages. Trained only on clean data and then activated at inference, the method improves accident detection rate over a vanilla attention baseline and a consensus defense by as much as 26 percent across jitter, dropout, local geometry attacks, and gradient-based PGD in three CARLA scenarios. A reader who cares about deployable multi-robot safety would care because the defense adds no new parameters and does not require attack-specific retraining.

What carries the argument

The plug-and-play reweighting module: after ordinary attention aggregation, compute Euclidean distances of neighbor value embeddings to the aggregated embedding, take the median and median absolute deviation of those distances, form a Tukey-style weight that is near 1 near the median and near 0 for large deviations, and re-normalize attention with those weights before the final sum.

What would settle it

Apply a coordinated attack that moves a majority of each critical neighborhood away from the true local structure (or that makes the true object itself a sparse outlier relative to the median) and check whether accident detection rate falls to or below the vanilla baseline on the same overtaking, left-turn, and red-light scenarios.

Watch

Extended reading notes

Core claim

The authors establish that a median-based, plug-and-play reweighting of attention weights, applied to neighborhood value embeddings both inside each vehicle's encoder and inside the ego decoder, produces resilient collaborative braking decisions under perceptual noise and five attack types without any additional training or architectural change, and that this yields state-of-the-art accident detection rates in high-fidelity connected-driving simulations.

Load-bearing premise

Under the attacks that matter, enough of each local neighborhood stays geometrically consistent that the median and its absolute deviation remain trustworthy references, so down-weighting outliers does not also erase sparse but decision-critical points.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes RCDM, an attention-based encoder–decoder for collaborative braking decisions among connected vehicles, augmented by a plug-and-play reweighting module. For each point, distances of neighbor value embeddings to the aggregated embedding yield a local median and MAD; neighbors far from that median receive Tukey-like weights that re-normalize attention (Eqs. 4–9, 12–13). Reweighting is deactivated in clean training and activated only at inference, with no extra parameters. In CARLA/AutoCastSim overtaking, left-turn, and red-light scenarios, under Jitter, Dropout, LNR, LRS, and 3-step PGD applied to ego, k collaborators, or all vehicles, RCDM improves Accident Detection Rate over a vanilla attention baseline and RoboSAC, with large gains under PGD-All (e.g., Overtaking 0.68→0.95).

Significance. If the gains hold under broader defenses and failure regimes, the work is practically useful: a training-free, architecture-agnostic robustness layer for attention-based collaborative perception is more deployable than attack-specific adversarial training or certified methods with restrictive threat models. Strengths include a clean train/execute separation, coherent robust-statistics motivation, multi-scenario multi-attack evaluation with corruption statistics (Table I), and attention-change visualizations (Fig. 4). The contribution is incremental relative to robust attention and consensus defenses, but the CAD-specific integration and plug-and-play framing are relevant to safety-critical multi-robot systems.

major comments (3)
  1. [Table II; §II.B; Abstract] The SOTA claim (Abstract; §I; Table II caption) rests on comparison to only Vanilla and RoboSAC. §II.B surveys adversarial training, certified defenses (PointGuard/PointCert-style), point removal/denoising, and other consensus methods, none of which appear as baselines. RoboSAC is also undefined (marked “—”) whenever the ego is attacked, so the comparison is incomplete precisely where RCDM’s ego-side reweighting is most relevant. Without at least one strong alternative from each major defense family on the same protocol, “outperforms existing methods by up to 26%” and “state-of-the-art” are not supported.
  2. [§III.B–C, Eqs. (4)–(8), (12)–(13)] The reweighting (Eqs. 4–8, 12–13) treats the median of {r_ij} over K=16 neighbors as a reliable local reference. This is load-bearing for both individual and collaborative resilience, yet the paper never reports neighborhood corruption fraction, MAD stability, or performance when a majority of neighbors are coordinated outliers (or when decision-critical structure is sparse). The evaluated attacks leave residual structure (Table I; Fig. 4), so Table II gains do not establish the claimed generality of plug-and-play resilience. A stress test or explicit failure-mode analysis is needed before the central claim can stand.
  3. [§III.B Eqs. (6)–(7); §IV.A] Free parameters k (Eq. 7), ε (Eq. 6), and neighborhood size K=16 are fixed without sensitivity or ablation. Attack intensities (σ=1.0 m, 60% dropout, LNR/LRS fractions, PGD ε/α/iters) are likewise fixed. Because the method’s only robustness mechanism is this reweighting, the reported ADR lifts could be tuned to this suite. At minimum, ablate k and K and report ADR under denser local corruption or higher PGD iteration counts.
minor comments (5)
  1. [§III.A–C] Notation switches between P′ / m′ and ˆP for the merged cloud; h_i is used both before and after reweighting. Unify symbols in §III.
  2. [Table II] Table II header text is garbled (“COMPARED TO OTHERSOTAMETHODS”). Clean for camera-ready.
  3. [Fig. 4; §IV.C] Fig. 4 Δ_total values are informative but lack a control (e.g., random reweighting) and a quantitative link to ADR; a short correlation or ROI-only ADR would strengthen the figure.
  4. [§IV.A; §V] Binary brake-only decisions and 12 train / 12 test trials are narrow; state this limitation more clearly in §V alongside the multi-robot and multi-modal extensions.
  5. [§II.B; [46]] Related work on robust transformers / ProTransformer [46] is cited but not contrasted with the median–MAD design; a sentence on differences would help position novelty.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: reweighting is a fixed robust-statistics formula trained only on clean data and evaluated on held-out corrupted trials; ADR gains are not forced by construction.

full rationale

RCDM’s load-bearing chain is algorithmic and empirical, not definitional. The encoder/decoder attentions (Eqs. 1–3, 11) are standard PointTransformer-style attention. The reweighting (Eqs. 4–8, 9–10, 12–13) is a fixed, parameter-light robust-statistics map (median distance, MAD, Tukey-like biweight) applied to already-trained attention at inference only; training deactivates reweighting and uses BCE on clean data (§III.D). Nothing in the weight formula is fitted to attack labels, ADR, or the reported gains. Evaluation is on held-out CARLA/AutoCastSim trials under five external attack models (Jitter, Dropout, LNR, LRS, PGD) with ADR as an independent metric (Table II). Self-citations of prior collaborative decision-making work ([19], [23]) appear only as related context and do not justify uniqueness or force the reweighting result; the method is motivated by external robust-statistics literature ([42]–[46]). There is no self-definitional loop, no fitted input renamed as prediction, no uniqueness theorem imported from the authors, and no renaming of a known empirical pattern as a derivation. Circularity score is therefore 0.

Assumptions & free parameters 4 free parameters · 4 assumptions · 1 invented entities

The central claim rests on standard attention fusion, classical robust location/scale (median, MAD), and domain assumptions about LiDAR local structure under the paper’s attack models. Free parameters are the reweighting sensitivity and floor, neighborhood size, and attack/simulation knobs that define the evaluation regime. No new physical entities are postulated; RCDM is an algorithmic construction.

free parameters (4)
  • reweighting sensitivity k (Eq. 7)
    Controls how quickly normalized deviations drive weights to zero; described as a hyperparameter with no reported selection procedure or sensitivity study.
  • MAD floor ε (Eq. 6)
    Lower bound to avoid division by zero; ad hoc numerical stabilizer affecting weight scale.
  • neighborhood size K=16
    Fixed k-NN size for local aggregation; shapes both attention and median statistics.
  • attack intensity settings (σ=1.0 m, 60% dropout, LNR/LRS fractions, PGD ε=1.0/α=0.4/3 iters)
    Define the corruption regime in which the ‘up to 26%’ and SOTA claims are measured; not learned but load-bearing for reported gains.
assumptions (4)
  • domain assumption Local neighborhoods of LiDAR points admit a consistent structure whose median distance in value-embedding space is a reliable reference for identifying corrupted neighbors.
    Core premise of Eqs. 4–8 and 12–13 in §III.B–C; without it reweighting can suppress true structure or fail under coordinated outliers.
  • domain assumption Training on clean collaborative data with standard attention yields embeddings for which inference-time reweighting improves decisions under unseen corruptions.
    Stated in §III.D; reweighting is never trained, so transfer of clean-trained attention to attacked inputs is assumed.
  • standard math Median and MAD are robust location/scale estimators for outlier down-weighting (classical robust statistics).
    Invoked via citations [42]–[46]; standard background, not proved in-paper.
  • domain assumption Collaborator points can be transformed into the ego frame via GNSS/HD map with sufficient accuracy for fusion.
    §III.A problem setup; localization error is not modeled in the attack suite.
invented entities (1)
  • RCDM plug-and-play reweighting module (median/MAD attention reweight)
    purpose: Down-weight corrupted neighbors in encoder and decoder attention without new learned parameters.
    Algorithmic construct, not a physical entity; independent_evidence false because validation is only the paper’s own CARLA ADR tables.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Plug-and-Play Reweighting for Resilient Collaborative Decision-Making in Connected Autonomous Driving." pith.science (2026). https://pith.science/paper/SSLNULBQ

@misc{pith2026260710037,
  author       = {Pith},
  title        = {Pith review of: Plug-and-Play Reweighting for Resilient Collaborative Decision-Making in Connected Autonomous Driving},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SSLNULBQ}},
  note         = {Machine review of arXiv:2607.10037}
}
read the original abstract

Collaborative decision-making is a fundamental capability in multi-robot systems, such as connected autonomous vehicles. However, perceptual noise and adversarial attacks in collaborators can severely affect decision reliability. Overall, existing methods typically rely on retraining with attack-specific defenses or on restrictive perturbation assumptions to improve resilience, which limits their practicality. In this paper, we propose a novel Resilient Collaborative Decision-Making (RCDM) framework that consists of an attention-based encoder for extracting individual robot perceptual embeddings and an attention-based decoder for fusing collaborator perceptions and making decisions. To improve resilience to corrupted observations, we design a novel plug-and-play reweighting module that down-weights the influence of corrupted inputs by analyzing the consistency of neighborhood points relative to the local structure and assigning smaller weights to points that deviate strongly from the local median. This module can be seamlessly integrated into attention-based collaborative decision-making without requiring additional training. We evaluate our method in high-fidelity simulations, considering perceptual noise and five types of attacks across diverse accident-prone scenarios. Experimental results demonstrate that our approach consistently outperforms existing methods by up to 26% and achieves state-of-the-art resilient performance.

Figures

Figures reproduced from arXiv: 2607.10037 by the authors.

Figure 1
Figure 1. An illustrative example of resilient collaborative decision-making in [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of our RCDM method. We employ an attention-based encoder-decoder architecture for collaborative decision-making. Due to the [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Illustration of scenarios and attacks in experiments. The top row [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: The visualization of attention changes of downsampled LiDAR [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

53 extracted references · 1 linked inside Pith

  1. [1]

    Multi-robot system for cooperative exploration in unknown environments: A Survey,

    C. Wang, C. Yu, X. Xu, Y . Gao, X. Yang, W. Tang, S. Yu, Y . Chen, F. Gao, Z. Jian,et al., “Multi-robot system for cooperative exploration in unknown environments: A Survey,”arXiv preprint arXiv:2503.07278, 2025

    arXiv 2025

  2. [2]

    Cooperative heterogeneous multi-robot systems: A Survey,

    Y . Rizk, M. Awad, and E. W. Tunstel, “Cooperative heterogeneous multi-robot systems: A Survey,”ACM Computing Surveys, vol. 52, 2019

    2019

  3. [3]

    Directed-cp: Directed collabo- rative perception for connected and autonomous vehicles via proactive attention,

    Y . Tao, S. Hu, Z. Fang, and Y . Fang, “Directed-cp: Directed collabo- rative perception for connected and autonomous vehicles via proactive attention,” inICRA, 2025

    2025

  4. [4]

    A mixed-reality- augmented deep reinforcement learning approach for multi-robot safe motion generation in human–robot collaborative manufacturing cells,

    C. Li, Y . Yin, H. Ye, P. Zheng, and S. K. Gupta, “A mixed-reality- augmented deep reinforcement learning approach for multi-robot safe motion generation in human–robot collaborative manufacturing cells,” IEEE Transactions on Automation Science and Engineering, vol. 22, pp. 21 033–21 046, 2025

    2025

  5. [5]

    Adversarial attacks on multi-agent communication,

    J. Tu, T. Wang, J. Wang, S. Manivasagam, M. Ren, and R. Urtasun, “Adversarial attacks on multi-agent communication,” inICCV, 2021

    2021

  6. [6]

    Art-point: Improving rotation ro- bustness of point cloud classifiers via adversarial rotation,

    R. Wang, Y . Yang, and D. Tao, “Art-point: Improving rotation ro- bustness of point cloud classifiers via adversarial rotation,” inCVPR, 2022

    2022

  7. [7]

    Robust com- municative multi-agent reinforcement learning with active defense,

    L. Yu, Y . Qiu, Q. Yao, Y . Shen, X. Zhang, and J. Wang, “Robust com- municative multi-agent reinforcement learning with active defense,” in AAAI, 2024

    2024

  8. [8]

    Pointguard: Provably robust 3D point cloud classification,

    H. Liu, J. Jia, and N. Z. Gong, “Pointguard: Provably robust 3D point cloud classification,” inCVPR, 2021

    2021

Show all 53 references
  1. [9]

    Pointcert: Point cloud classification with deterministic certified robustness guarantees,

    J. Zhang, J. Jia, H. Liu, and N. Z. Gong, “Pointcert: Point cloud classification with deterministic certified robustness guarantees,” in CVPR, 2023

    2023

  2. [10]

    Dup- net: Denoiser and upsampler network for 3D adversarial point clouds defense,

    H. Zhou, K. Chen, W. Zhang, H. Fang, W. Zhou, and N. Yu, “Dup- net: Denoiser and upsampler network for 3D adversarial point clouds defense,” inICCV, 2019

    2019

  3. [11]

    Adversarially robust 3D point cloud recognition using self- supervisions,

    J. Sun, Y . Cao, C. Choy, Z. Yu, A. Anandkumar, Z. M. Mao, and C. Xiao, “Adversarially robust 3D point cloud recognition using self- supervisions,” inNeurIPS, 2021

    2021

  4. [12]

    Among us: Adversarially robust collaborative perception by consensus,

    Y . Li, Q. Fang, J. Bai, S. Chen, F. Juefei-Xu, and C. Feng, “Among us: Adversarially robust collaborative perception by consensus,” inICCV, 2023

    2023

  5. [13]

    Robust consensus for multi-agent systems commu- nicating over stochastic uncertain networks,

    Z. Li and J. Chen, “Robust consensus for multi-agent systems commu- nicating over stochastic uncertain networks,”SIAM Journal on Control and Optimization, vol. 57, pp. 3553–3570, 2019

    2019

  6. [14]

    De- centralized cooperative lane-changing decision-making for connected autonomous vehicles,

    J. Nie, J. Zhang, W. Ding, X. Wan, X. Chen, and B. Ran, “De- centralized cooperative lane-changing decision-making for connected autonomous vehicles,”IEEE access, vol. 4, pp. 9413–9420, 2017

    2017

  7. [15]

    A rule-based cooperative merg- ing strategy for connected and automated vehicles,

    J. Ding, L. Li, H. Peng, and Y . Zhang, “A rule-based cooperative merg- ing strategy for connected and automated vehicles,”IEEE Transactions on Intelligent Transportation Systems, vol. 21, no. 8, pp. 3436–3446, 2019

    2019

  8. [16]

    Cooperative deci- sion making of connected automated vehicles at multi-lane merging zone: A coalitional game approach,

    P. Hang, C. Lv, C. Huang, Y . Xing, and Z. Hu, “Cooperative deci- sion making of connected automated vehicles at multi-lane merging zone: A coalitional game approach,”IEEE Transactions on Intelligent Transportation Systems, vol. 23, no. 4, pp. 3829–3841, 2021

    2021

  9. [17]

    Decision making of connected automated vehicles at an unsignalized roundabout consider- ing personalized driving behaviours,

    P. Hang, C. Huang, Z. Hu, Y . Xing, and C. Lv, “Decision making of connected automated vehicles at an unsignalized roundabout consider- ing personalized driving behaviours,”IEEE Transactions on Vehicular Technology, vol. 70, no. 5, pp. 4051–4064, 2021

    2021

  10. [18]

    Graph neural network and reinforcement learning for multi-agent cooperative control of con- nected autonomous vehicles,

    S. Chen, J. Dong, P. Ha, Y . Li, and S. Labi, “Graph neural network and reinforcement learning for multi-agent cooperative control of con- nected autonomous vehicles,”Computer-Aided Civil and Infrastructure Engineering, vol. 36, no. 7, pp. 838–857, 2021

    2021

  11. [19]

    Collaborative decision-making using spatiotemporal graphs in connected autonomy,

    P. Gao, Y . Shen, and M. C. Lin, “Collaborative decision-making using spatiotemporal graphs in connected autonomy,” inICRA, 2024

    2024

  12. [20]

    Cooperative behavior plan- ning for automated driving using graph neural networks,

    M. Klimke, B. V ¨olz, and M. Buchholz, “Cooperative behavior plan- ning for automated driving using graph neural networks,” in2022 IEEE Intelligent Vehicles Symposium (IV). IEEE, 2022, pp. 167– 174

    2022

  13. [21]

    Coopernaut: End-to- end driving with cooperative perception for networked vehicles,

    J. Cui, H. Qiu, D. Chen, P. Stone, and Y . Zhu, “Coopernaut: End-to- end driving with cooperative perception for networked vehicles,” in CVPR, 2022

    2022

  14. [22]

    End- to-end autonomous driving through V2X cooperation,

    H. Yu, W. Yang, J. Zhong, Z. Yang, S. Fan, P. Luo, and Z. Nie, “End- to-end autonomous driving through V2X cooperation,” inAAAI, 2025

    2025

  15. [23]

    MMCD: Multi-modal collaborative decision-making for connected autonomy with knowledge distillation,

    R. Liu, Z. Wang, P. Gao, Y . Shen, P. Tokekar, and M. Lin, “MMCD: Multi-modal collaborative decision-making for connected autonomy with knowledge distillation,” inIROS, 2025

    2025

  16. [24]

    Who2com: Collaborative perception via learnable handshake com- munication,

    Y .-C. Liu, J. Tian, C.-Y . Ma, N. Glaser, C.-W. Kuo, and Z. Kira, “Who2com: Collaborative perception via learnable handshake com- munication,” inICRA, 2020

    2020

  17. [25]

    When2com: Multi-agent perception via communication graph grouping,

    Y .-C. Liu, J. Tian, N. Glaser, and Z. Kira, “When2com: Multi-agent perception via communication graph grouping,” inCVPR, 2020

    2020

  18. [26]

    Where2comm: Communication-efficient collaborative perception via spatial confi- dence maps,

    Y . Hu, S. Fang, Z. Lei, Y . Zhong, and S. Chen, “Where2comm: Communication-efficient collaborative perception via spatial confi- dence maps,” inNeurIPS, 2022

    2022

  19. [27]

    Generating 3D adversarial point clouds,

    C. Xiang, C. R. Qi, and B. Li, “Generating 3D adversarial point clouds,” inCVPR, 2019

    2019

  20. [28]

    Pointcloud saliency maps,

    T. Zheng, C. Chen, J. Yuan, B. Li, and K. Ren, “Pointcloud saliency maps,” inICCV, 2018

    2018

  21. [29]

    On isometry robustness of deep 3D point cloud models under adversarial attacks,

    Y . Zhao, Y . Wu, C. Chen, and A. Lim, “On isometry robustness of deep 3D point cloud models under adversarial attacks,” inCVPR, 2020

    2020

  22. [30]

    Adversarial sensor attack on LiDAR-based perception in autonomous driving,

    Y . Cao, C. Xiao, B. Cyr, Y . Zhou, W. Park, S. Rampazzi, Q. A. Chen, K. Fu, and Z. M. Mao, “Adversarial sensor attack on LiDAR-based perception in autonomous driving,” inCCS, 2019

    2019

  23. [31]

    Fooling LiDAR perception via adversarial trajectory perturbation,

    Y . Li, C. Wen, F. Juefei-Xu, and C. Feng, “Fooling LiDAR perception via adversarial trajectory perturbation,” inICCV, 2021

    2021

  24. [32]

    Security analysis of Camera-LiDAR fusion against Black-Box attacks on autonomous vehicles,

    R. S. Hallyburton, Y . Liu, Y . Cao, Z. M. Mao, and M. Pajic, “Security analysis of Camera-LiDAR fusion against Black-Box attacks on autonomous vehicles,” inUSENIX Security, 2022

    2022

  25. [33]

    Towards universal physical attacks on cascaded Camera-LiDAR 3D object detection models,

    M. Abdelfattah, K. Yuan, Z. J. Wang, and R. Ward, “Towards universal physical attacks on cascaded Camera-LiDAR 3D object detection models,” inICIP, 2021

    2021

  26. [34]

    A backdoor attack against 3D point cloud classifiers,

    Z. Xiang, D. J. Miller, S. Chen, X. Li, and G. Kesidis, “A backdoor attack against 3D point cloud classifiers,” inICCV, 2021

    2021

  27. [35]

    Towards backdoor attacks against LiDAR object detection in autonomous driving,

    Y . Zhang, Y . Zhu, Z. Liu, C. Miao, F. Hajiaghajani, L. Su, and C. Qiao, “Towards backdoor attacks against LiDAR object detection in autonomous driving,” inSenSys, 2022

    2022

  28. [36]

    Badfusion: 2D-Oriented backdoor attacks against 3D object detection,

    S. S. Chaturvedi, L. Zhang, W. Zhang, P. He, and X. Yuan, “Badfusion: 2D-Oriented backdoor attacks against 3D object detection,” inIJCAI, 2024

    2024

  29. [37]

    Improving adversarial robustness of 3D point cloud classification models,

    G. Li, G. Xu, H. Qiu, R. He, J. Li, and T. Zhang, “Improving adversarial robustness of 3D point cloud classification models,” in ECCV, 2022

    2022

  30. [38]

    Benchmarking and analyzing robust point cloud recognition: Bag of tricks for defending adversarial examples,

    Q. Ji, L. Wang, C. Shi, S. Hu, Y . Chen, and L. Sun, “Benchmarking and analyzing robust point cloud recognition: Bag of tricks for defending adversarial examples,” inICCV, 2023

    2023

  31. [39]

    Commit: certifying robustness of multi-sensor fusion systems against semantic attacks,

    Z. Huang, W. Chu, L. Li, C. Xu, and B. Li, “Commit: certifying robustness of multi-sensor fusion systems against semantic attacks,” inAAAI, 2025

    2025

  32. [40]

    Pointcvar: risk-optimized outlier removal for robust 3d point cloud classification,

    X. Li, J. Lu, H. Ding, C. Sun, J. T. Zhou, and Y . M. Chee, “Pointcvar: risk-optimized outlier removal for robust 3d point cloud classification,” inAAAI, 2024

    2024

  33. [41]

    Mis-spoke or mis-lead: Achieving robustness in multi-agent communicative reinforcement learning,

    W. Xue, W. Qiu, B. An, Z. Rabinovich, S. Obraztsova, and C. K. Yeo, “Mis-spoke or mis-lead: Achieving robustness in multi-agent communicative reinforcement learning,” inAAMAS, 2022

    2022

  34. [42]

    The fitting of power series, meaning polynomials, illustrated on band-spectroscopic data,

    A. E. Beaton and J. W. Tukey, “The fitting of power series, meaning polynomials, illustrated on band-spectroscopic data,”Technometrics, vol. 16, pp. 147–185, 1974

    1974

  35. [43]

    Bloomfield and W

    P. Bloomfield and W. L. Steiger,Least absolute deviations: theory, applications, and algorithms. Springer, 1983, vol. 6

    1983

  36. [44]

    Robust regression: asymptotics, conjectures and monte carlo,

    P. J. Huber, “Robust regression: asymptotics, conjectures and monte carlo,”The Annals of Statistics, vol. 1, pp. 799–821, 1973

    1973

  37. [45]

    Nearly unbiased variable selection under minimax concave penalty,

    C.-H. Zhang, “Nearly unbiased variable selection under minimax concave penalty,”The Annals of Statistics, vol. 38, pp. 894–942, 2010

    2010

  38. [46]

    Protransformer: Robustify transformers via plug-and-play paradigm,

    Z. Hou, W. Gao, Y . Shen, F. Wang, and X. Liu, “Protransformer: Robustify transformers via plug-and-play paradigm,”NeurIPS, 2024

    2024

  39. [47]

    CARLA: An open urban driving simulator,

    A. Dosovitskiy, G. Ros, F. Codevilla, A. Lopez, and V . Koltun, “CARLA: An open urban driving simulator,” inCoRL, 2017

    2017

  40. [48]

    Autocast: Scalable infrastructure-less cooperative perception for distributed collaborative driving,

    H. Qiu, P. Huang, N. Asavisanu, X. Liu, K. Psounis, and R. Govin- dan, “Autocast: Scalable infrastructure-less cooperative perception for distributed collaborative driving,” inMobiSys, 2022

    2022

  41. [49]

    Benchmarking and analyzing point cloud classification under corruptions,

    J. Ren, L. Pan, and Z. Liu, “Benchmarking and analyzing point cloud classification under corruptions,” inICML, 2022

    2022

  42. [50]

    Explicitly perceiving and preserving the local geometric structures for 3D point cloud attack,

    D. Liu and W. Hu, “Explicitly perceiving and preserving the local geometric structures for 3D point cloud attack,” inAAAI, 2024

    2024

  43. [51]

    Towards deep learning models resistant to adversarial attacks,

    A. Madry, A. Makelov, L. Schmidt, D. Tsipras, and A. Vladu, “Towards deep learning models resistant to adversarial attacks,” in ICLR, 2018

    2018

  44. [52]

    Point transformer,

    H. Zhao, L. Jiang, J. Jia, P. H. Torr, and V . Koltun, “Point transformer,” inICCV, 2021

    2021

  45. [53]

    Adam: A method for stochastic optimiza- tion,

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimiza- tion,” inICLR, 2015

    2015

Pith tools

Reviewed July 14, 2026 · model on record in the stance chip above.