FedADAS: Communication-Efficient Federated Distillation for On-Device Driver Yawn Recognition in Vehicular Networks

Ahmed Mujtaba; Gleb Radchenko; Marc Masana; Radu Prodan

arxiv: 2605.19480 · v1 · pith:ZR5EOYGXnew · submitted 2026-05-19 · 💻 cs.DC

FedADAS: Communication-Efficient Federated Distillation for On-Device Driver Yawn Recognition in Vehicular Networks

Ahmed Mujtaba , Gleb Radchenko , Marc Masana , Radu Prodan This is my paper

Pith reviewed 2026-05-20 02:31 UTC · model grok-4.3

classification 💻 cs.DC

keywords federated distillationdriver yawn recognitionvehicular networksmodel heterogeneitycommunication efficiencyedge computingpersonalized learningfederated learning

0 comments

The pith

FedADAS enables full model heterogeneity for driver yawn recognition by exchanging only soft logits on a shared public dataset.

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

The paper introduces FedADAS as a federated distillation framework for collaborative on-device learning of driver yawn recognition models across vehicular networks. Standard federated learning requires identical model architectures and sends large parameter updates, which creates high communication costs and prevents customization to each vehicle's hardware. FedADAS instead has clients exchange soft predictions computed on one common public dataset, which supports arbitrary model sizes per vehicle while still allowing knowledge to transfer. A sympathetic reader would care because this makes privacy-preserving, hardware-aware training practical for safety-critical applications like fatigue detection in real driving conditions with diverse data and devices.

Core claim

FedADAS is a federated distillation method that achieves full model heterogeneity by exchanging only soft logits on a shared public dataset rather than model parameters or gradients. This lets each vehicle run a customized architecture suited to its computational constraints while handling extreme data heterogeneity across clients. In tests with up to 115 edge devices, the approach outperforms conventional federated learning at higher participation rates and delivers up to 9974x lower communication volume while preserving a strong balance between personalization and generalization. Two supporting architectures are supplied: a performance-efficient model reaching 98.3 percent F1-score and a 0

What carries the argument

Soft-logit exchange on a shared public dataset within a federated distillation protocol; the mechanism transfers knowledge across clients without forcing uniform model architectures or large parameter shipments.

If this is right

Vehicles can deploy models sized exactly to their available memory and compute, from full performance models down to tiny ones that train on edge hardware.
Communication volume drops far enough to allow frequent model updates even in large fleets without saturating network links.
Personalization improves for individual drivers while generalization across the fleet stays competitive under non-identical data.
Both training and inference become feasible directly on resource-limited onboard processors such as Jetson devices.

Where Pith is reading between the lines

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

The same soft-logit distillation pattern could be tried for other vehicle perception tasks such as distraction or pedestrian detection.
When a suitable public dataset is hard to obtain, synthetic data generation might serve as a practical substitute for the knowledge transfer step.
Mixed fleets containing both high-end and low-end vehicles would provide a direct test of how well the personalization-generalization tradeoff holds at scale.

Load-bearing premise

The framework depends on one shared public dataset being representative enough for soft logits to transfer useful knowledge across clients that have very different private data distributions and device capabilities.

What would settle it

Replacing the public dataset with one that poorly matches the clients' data distributions and checking whether the accuracy and communication advantages over standard federated learning disappear would test the central claim.

Figures

Figures reproduced from arXiv: 2605.19480 by Ahmed Mujtaba, Gleb Radchenko, Marc Masana, Radu Prodan.

**Figure 2.** Figure 2: DMS pipeline for driver yawn recognition using DL models on edge de [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

**Figure 3.** Figure 3: Training and inference efficiency scores of the DMS yawn recognition [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗

read the original abstract

Driver fatigue is a critical safety concern in advanced driver assistance systems. Driver monitoring models trained off-site on static datasets adapt poorly to real-world conditions, while standard federated learning imposes high communication overhead, assumes homogeneous architectures, and struggles with personalized driver data. We present FedADAS, a federated distillation framework enabling collaborative on-device learning across heterogeneous vehicular networks. FedADAS enables full model heterogeneity by exchanging only soft logits on a shared public dataset, allowing each vehicle to run a customized model tailored to its computational constraints. Additionally, we introduce a yawn recognition pipeline supporting training and inference on edge devices that provides two robust architectures: Performance-Efficient (99.7 MB) achieving 98.3% F1-score with 1.99ms inference time on a Jetson NANO, and a Memory-Efficient (0.6 MB) that trains an epoch in 6.12 minutes on a Jetson AGX Orin. In experiments with up to 115 edge clients, FedADAS significantly outperforms traditional federated learning approaches at higher client participation, achieving up to 9974x reduction in communication cost while maintaining a superior tradeoff between personalization and generalization under extreme data heterogeneity, demonstrating its suitability for real-world deployment. Code is available at https://opensource.silicon-austria.com/mujtabaa/fedadas

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

FedADAS shows how federated distillation lets heterogeneous vehicle models share only logits on a public set, cutting communication sharply in 115-client tests while hitting good F1 on edge hardware.

read the letter

The main point is that this paper takes federated distillation and applies it to on-device yawn detection across cars with very different hardware and data. Clients keep their own model sizes and only exchange soft predictions on one shared public dataset, which lets them avoid shipping full weights. In the reported runs that reaches up to 9974x lower communication cost than standard federated averaging while still showing a better personalization-generalization balance under the claimed extreme heterogeneity. They also ship two concrete architectures: a 99.7 MB performance version that runs at 1.99 ms on Jetson Nano and a 0.6 MB memory version that trains an epoch in roughly six minutes on AGX Orin, both reaching around 98 % F1 in their tests. Code is released, which helps reproducibility. That combination of application focus, measured device numbers, and the scale of the communication claim is the useful incremental piece. The experiments with 115 clients and higher participation rates are concrete enough to be worth looking at. The soft spot is the reliance on a single fixed public dataset for the distillation step. The abstract and setup describe extreme client heterogeneity but do not appear to include direct measures of how well that public set overlaps with the local distributions, such as label-shift statistics or distance metrics. If the public data is not representative, both the reported accuracy and the communication savings rest partly on that choice rather than purely on the distillation mechanism. The baselines are traditional federated learning, which is fair, but fuller details on statistical tests and data splits would strengthen the claims. This is aimed at applied researchers working on edge AI for vehicles or on communication-efficient federated methods for heterogeneous devices. It has enough empirical grounding and practical framing to deserve a serious referee, even if reviewers will want to press on the public-dataset assumption and the exact experimental controls.

Referee Report

1 major / 2 minor

Summary. The manuscript presents FedADAS, a federated distillation framework for on-device driver yawn recognition in vehicular networks. It enables full model heterogeneity by exchanging only soft logits on a shared public dataset, allowing each client to use a customized architecture. Experiments with up to 115 clients under extreme data heterogeneity report up to 9974x communication cost reduction, 98.3% F1-score on a Performance-Efficient model (99.7 MB, 1.99 ms inference on Jetson NANO), and a Memory-Efficient model (0.6 MB) while outperforming traditional federated learning; code is released.

Significance. If the empirical claims hold, the work offers a practical route to personalized, communication-efficient federated learning for resource-constrained vehicular edge devices, addressing key limitations of standard FL in heterogeneous settings. The explicit code release supports reproducibility and is a clear strength.

major comments (1)

[Section 4 / experimental setup] Section 4 / experimental setup: The central claims of superior personalization-generalization tradeoff and 9974x communication reduction rest on the assumption that soft-logit distillation on one fixed public dataset transfers useful knowledge across clients with sharply heterogeneous yawn data distributions. No quantitative measures (e.g., Wasserstein distance, label-shift statistics, or coverage of the convex hull of client distributions) are reported to validate that the public set is representative; this is load-bearing for the reported gains.

minor comments (2)

[Abstract] Abstract: '9974x' should be rendered as '9974×' for typographic consistency with scientific notation.
[Abstract] Abstract: The two architectures are introduced with concrete metrics; adding explicit forward references to the sections that define their layer counts, training procedures, and exact communication-volume calculations would aid readers.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive comment on the experimental validation. We address the point below and will revise the manuscript accordingly.

read point-by-point responses

Referee: The central claims of superior personalization-generalization tradeoff and 9974x communication reduction rest on the assumption that soft-logit distillation on one fixed public dataset transfers useful knowledge across clients with sharply heterogeneous yawn data distributions. No quantitative measures (e.g., Wasserstein distance, label-shift statistics, or coverage of the convex hull of client distributions) are reported to validate that the public set is representative; this is load-bearing for the reported gains.

Authors: We agree that explicit quantitative measures of distributional similarity would strengthen the claims. The public dataset was assembled from established yawn and driver monitoring benchmarks selected for broad coverage of expressions, lighting, and demographics, and the experiments demonstrate effective knowledge transfer under the induced extreme heterogeneity. To directly address the concern, the revised manuscript will add Wasserstein distance computations on extracted features, label-shift statistics, and a simple coverage analysis of the public set relative to the client partitions. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical framework with independent experimental validation

full rationale

The paper presents FedADAS as a practical federated distillation method that exchanges soft logits on a shared public dataset to support model heterogeneity. All reported outcomes (communication cost reductions up to 9974x, F1-scores, inference times, and personalization-generalization tradeoffs) are obtained directly from simulations with up to 115 heterogeneous clients and two edge architectures; these quantities are measured quantities, not quantities derived from equations that reference the same quantities by construction. No self-definitional steps, fitted-input predictions, or load-bearing self-citations appear in the method description or results. The central claims rest on the experimental comparison to baseline federated learning rather than on any internal reduction or imported uniqueness theorem.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract provides limited technical detail; primary assumptions concern data availability and heterogeneity handling rather than explicit free parameters or new entities.

axioms (1)

domain assumption A shared public dataset is available and representative enough to support effective distillation across heterogeneous clients
Central to the method of exchanging soft logits without model weights.

pith-pipeline@v0.9.0 · 5782 in / 1287 out tokens · 52253 ms · 2026-05-20T02:31:10.837944+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

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

IndisputableMonolith/Foundation/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

FedADAS enables full model heterogeneity by exchanging only soft logits on a shared public dataset... KL divergence between the model's predictions and the ensemble soft labels
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

up to 9974x reduction in communication cost while maintaining a superior tradeoff between personalization and generalization under extreme data heterogeneity

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

34 extracted references · 34 canonical work pages · 1 internal anchor

[1]

In: Proceedings of the 5th ACM multimedia systems conference

Abtahi, S., Omidyeganeh, M., Shirmohammadi, S., Hariri, B.: Yawdd: A yawning detection dataset. In: Proceedings of the 5th ACM multimedia systems conference. pp. 24–28 (2014)

work page 2014
[2]

Administration, N.H.T.S.: Drowsy driving, https://www.nhtsa.gov/risky-driving/ drowsy-driving

work page
[3]

Journal of Trans- portation Engineering, Part A: Systems151(3), 04024126 (2025)

Al-Mahbashi, M., Li, G., Peng, Y., Al-Soswa, M., Debsi, A.: Real-time distracted driving detection based on gm-yolov8 on embedded systems. Journal of Trans- portation Engineering, Part A: Systems151(3), 04024126 (2025)

work page 2025
[4]

IEEE Transactions on Cybernetics52(12), 13821–13833 (2021)

Bai, J., Yu, W., Xiao, Z., Havyarimana, V., Regan, A.C., Jiang, H., Jiao, L.: Two-stream spatial–temporal graph convolutional networks for driver drowsiness detection. IEEE Transactions on Cybernetics52(12), 13821–13833 (2021)

work page 2021
[5]

ACM Transactions on In- telligent Systems and Technology15(3), 1–27 (2024)

Bano, S., Tonellotto, N., Cassarà, P., Gotta, A.: Fedcmd: A federated cross-modal knowledge distillation for drivers’ emotion recognition. ACM Transactions on In- telligent Systems and Technology15(3), 1–27 (2024)

work page 2024
[6]

Microprocessors and Microsystems99(2023)

Civik, E., Yuzgec, U.: Real-time driver fatigue detection system with deep learning on a low-cost embedded system. Microprocessors and Microsystems99(2023)

work page 2023
[7]

Scientific Reports 14(1), 25029 (2024)

Eid Kishawy, M.M., Abd El-Hafez, M.T., Yousri, R., Darweesh, M.S.: Federated learning system on autonomous vehicles for lane segmentation. Scientific Reports 14(1), 25029 (2024)

work page 2024
[8]

Sensors 25, 812 (2025)

Essahraui, S., Lamaakal, I., El Hamly, I., Maleh, Y., Ouahbi, I., El Makkaoui, K., Filali Bouami, M., Pławiak, P., Alfarraj, O., Abd El-Latif, A.A.: Real-time driver drowsiness detection using facial analysis and machine learning techniques. Sensors 25, 812 (2025)

work page 2025
[9]

IFAC-PapersOnLine53(2), 15374–15379 (2020)

He, H., Zhang, X., Jiang, F., Wang, C., Yang, Y., Liu, W., Peng, J.: A real-time driver fatigue detection method based on two-stage convolutional neural network. IFAC-PapersOnLine53(2), 15374–15379 (2020)

work page 2020
[10]

In: Proceedings of Neural Information Processing Systems Workshop (2014)

Hinton, G., Vinyals, O., Dean, J.: Distilling the knowledge in a neural network. In: Proceedings of Neural Information Processing Systems Workshop (2014)

work page 2014
[11]

IEEE Transactions on Vehicular Technology (2025)

Huang, X., Li, W., Liang, C., Cao, B., Zhou, M.: Environment-aware personalized heterogeneous federated distillation for dual-layer blockchain-enabled internet of vehicles. IEEE Transactions on Vehicular Technology (2025)

work page 2025
[12]

YOLOv11: An Overview of the Key Architectural Enhancements

Khanam, R., Hussain, M.: Yolov11: An overview of the key architectural enhance- ments. arXiv preprint arXiv:2410.17725 (2024)

work page internal anchor Pith review Pith/arXiv arXiv 2024
[13]

Kuo, J., Wang, X., Li, X., Xu, T.: Federated learning in autonomous ve- hicles using cross-border training (2023), https://developer.nvidia.com/blog/ federated-learning-in-autonomous-vehicles-using-cross-border-training/

work page 2023
[14]

IEEE Transactions on Machine Learning in Communications and Networking1, 210–224 (2023)

Lan, G., Liu, X.Y., Zhang, Y., Wang, X.: Communication-efficient federated learn- ing for resource-constrained edge devices. IEEE Transactions on Machine Learning in Communications and Networking1, 210–224 (2023)

work page 2023
[15]

In: Proceedings of Neural Information Processing Systems, FLDPC Workshop (2019)

Li, D., Wang, J.: Fedmd: Heterogenous federated learning via model distillation. In: Proceedings of Neural Information Processing Systems, FLDPC Workshop (2019)

work page 2019
[16]

In: 2024 International Joint Confer- ence on Neural Networks (IJCNN)

Liang, J., Li, J., Zhang, J., Zang, T.: Federated learning with data-free distillation for heterogeneity-aware autonomous driving. In: 2024 International Joint Confer- ence on Neural Networks (IJCNN). pp. 1–7. IEEE (2024)

work page 2024
[17]

Computer Networks (2025)

Liao, G., Yang, Y., Feng, Z.: Personalized federated learning through self- knowledge distillation in vehicular edge computing. Computer Networks (2025)

work page 2025
[18]

In: Proceedings of the European conference on computer vision (ECCV)

Ma, N., Zhang, X., Zheng, H.T., Sun, J.: Shufflenet v2: Practical guidelines for efficient cnn architecture design. In: Proceedings of the European conference on computer vision (ECCV). pp. 116–131 (2018) FedADAS 15

work page 2018
[19]

Sensors 23(21), 8741 (2023)

Majeed, F., Shafique, U., Safran, M., Alfarhood, S., Ashraf, I.: Detection of drowsi- ness among drivers using novel deep convolutional neural network model. Sensors 23(21), 8741 (2023)

work page 2023
[20]

In: Artificial intelligence and statistics

McMahan, B., Moore, E., Ramage, D., Hampson, S., y Arcas, B.A.: Communication-efficient learning of deep networks from decentralized data. In: Artificial intelligence and statistics. pp. 1273–1282. PMLR (2017)

work page 2017
[21]

In: Proceedings of the IEEE International Conference on Image Processing (2026)

Mujtaba, A., Radchenko, G., Masana, M., Prodan, R.: Yawdd+: Frame-level an- notations for accurate yawn prediction. In: Proceedings of the IEEE International Conference on Image Processing (2026)

work page 2026
[22]

In: 2025 3rd Interna- tional Conference on Federated Learning Technologies and Applications (FLTA)

Mujtaba, A., Radchenko, G., Prodan, R., Masana, M.: Federated distillation on edge devices: Efficient client-side filtering for non-iid data. In: 2025 3rd Interna- tional Conference on Federated Learning Technologies and Applications (FLTA). pp. 228–235 (2025)

work page 2025
[23]

International Journal of Intelligent Systems2025(1), 7406934 (2025)

Qin, L., Zhu, T., Zhou, W., Yu, P.S.: Knowledge distillation in federated learning: A survey on long lasting challenges and new solutions. International Journal of Intelligent Systems2025(1), 7406934 (2025)

work page 2025
[24]

Riya, F.F., Hoque, S., Zhao, X., Sun, J.S.: Smart driver monitoring robotic system to enhance road safety : A comprehensive review (2024)

work page 2024
[25]

IEEE Internet of Things Journal10(13), 11643–11654 (2023)

Shang, E., Liu, H., Yang, Z., Du, J., Ge, Y.: Fedbikd: Federated bidirectional knowledge distillation for distracted driving detection. IEEE Internet of Things Journal10(13), 11643–11654 (2023)

work page 2023
[26]

Nature Communications15(1), 349 (2024)

Shao, J., Wu, F., Zhang, J.: Selective knowledge sharing for privacy-preserving federated distillation without a good teacher. Nature Communications15(1), 349 (2024)

work page 2024
[27]

In: 2024 International conference on advances in data engineering and intelligent computing systems (ADICS)

Varghese, R., Sambath, M.: Yolov8: A novel object detection algorithm with en- hanced performance and robustness. In: 2024 International conference on advances in data engineering and intelligent computing systems (ADICS). pp. 1–6. IEEE (2024)

work page 2024
[28]

In: 2024 IEEE Interna- tional Conference on Big Data (BigData)

Wang, J., Gao, J.: Federated learning with knowledge distillation to mitigate catas- trophic forgetting and data heterogeneity in iov systems. In: 2024 IEEE Interna- tional Conference on Big Data (BigData). pp. 2914–2923. IEEE (2024)

work page 2024
[29]

Wang, X., Tang, Z., Guo, J., Meng, T., Wang, C., Wang, T., Jia, W.: Empowering edgeintelligence:Acomprehensivesurveyonon-deviceaimodels.ACMComputing Surveys57(9), 1–39 (2025)

work page 2025
[30]

IEEE Transac- tions on Vehicular Technology (2025)

Xiao, S., Huang, X., Zhou, M., Liang, C., Chen, Q.: Feddld: Dual-level federated distillation with adaptive knowledge transfer for dag-secured iovs. IEEE Transac- tions on Vehicular Technology (2025)

work page 2025
[31]

IEICE Transactions on Information and Systems p

XU, M., ZHAN, A., WU, C., WANG, Z.: A novel driver fatigue detection method based on dual-stream swin-transformer. IEICE Transactions on Information and Systems p. 2024EDL8094 (2025)

work page 2025
[32]

In: International Conference on Learning Representations (2023)

Zhang, J., Chen, C., Lyu, L.: Ideal: Query-efficient data-free learning from black- box models. In: International Conference on Learning Representations (2023)

work page 2023
[33]

In: Proc

Zhang, Y., Jia, R., Pei, H., Wang, W., Li, B., Song, D.: The secret revealer: Generative model-inversion attacks against deep neural networks. In: Proc. of the IEEE/CVF conference on computer vision and pattern recognition. pp. 253–261 (2020)

work page 2020
[34]

In: 2021 33rd Chinese Control and Decision Confer- ence (CCDC)

Zhou, C., Li, J.: A real-time driver fatigue monitoring system based on lightweight convolutional neural network. In: 2021 33rd Chinese Control and Decision Confer- ence (CCDC). pp. 1548–1553. IEEE (2021)

work page 2021

[1] [1]

In: Proceedings of the 5th ACM multimedia systems conference

Abtahi, S., Omidyeganeh, M., Shirmohammadi, S., Hariri, B.: Yawdd: A yawning detection dataset. In: Proceedings of the 5th ACM multimedia systems conference. pp. 24–28 (2014)

work page 2014

[2] [2]

Administration, N.H.T.S.: Drowsy driving, https://www.nhtsa.gov/risky-driving/ drowsy-driving

work page

[3] [3]

Journal of Trans- portation Engineering, Part A: Systems151(3), 04024126 (2025)

Al-Mahbashi, M., Li, G., Peng, Y., Al-Soswa, M., Debsi, A.: Real-time distracted driving detection based on gm-yolov8 on embedded systems. Journal of Trans- portation Engineering, Part A: Systems151(3), 04024126 (2025)

work page 2025

[4] [4]

IEEE Transactions on Cybernetics52(12), 13821–13833 (2021)

Bai, J., Yu, W., Xiao, Z., Havyarimana, V., Regan, A.C., Jiang, H., Jiao, L.: Two-stream spatial–temporal graph convolutional networks for driver drowsiness detection. IEEE Transactions on Cybernetics52(12), 13821–13833 (2021)

work page 2021

[5] [5]

ACM Transactions on In- telligent Systems and Technology15(3), 1–27 (2024)

Bano, S., Tonellotto, N., Cassarà, P., Gotta, A.: Fedcmd: A federated cross-modal knowledge distillation for drivers’ emotion recognition. ACM Transactions on In- telligent Systems and Technology15(3), 1–27 (2024)

work page 2024

[6] [6]

Microprocessors and Microsystems99(2023)

Civik, E., Yuzgec, U.: Real-time driver fatigue detection system with deep learning on a low-cost embedded system. Microprocessors and Microsystems99(2023)

work page 2023

[7] [7]

Scientific Reports 14(1), 25029 (2024)

Eid Kishawy, M.M., Abd El-Hafez, M.T., Yousri, R., Darweesh, M.S.: Federated learning system on autonomous vehicles for lane segmentation. Scientific Reports 14(1), 25029 (2024)

work page 2024

[8] [8]

Sensors 25, 812 (2025)

Essahraui, S., Lamaakal, I., El Hamly, I., Maleh, Y., Ouahbi, I., El Makkaoui, K., Filali Bouami, M., Pławiak, P., Alfarraj, O., Abd El-Latif, A.A.: Real-time driver drowsiness detection using facial analysis and machine learning techniques. Sensors 25, 812 (2025)

work page 2025

[9] [9]

IFAC-PapersOnLine53(2), 15374–15379 (2020)

He, H., Zhang, X., Jiang, F., Wang, C., Yang, Y., Liu, W., Peng, J.: A real-time driver fatigue detection method based on two-stage convolutional neural network. IFAC-PapersOnLine53(2), 15374–15379 (2020)

work page 2020

[10] [10]

In: Proceedings of Neural Information Processing Systems Workshop (2014)

Hinton, G., Vinyals, O., Dean, J.: Distilling the knowledge in a neural network. In: Proceedings of Neural Information Processing Systems Workshop (2014)

work page 2014

[11] [11]

IEEE Transactions on Vehicular Technology (2025)

Huang, X., Li, W., Liang, C., Cao, B., Zhou, M.: Environment-aware personalized heterogeneous federated distillation for dual-layer blockchain-enabled internet of vehicles. IEEE Transactions on Vehicular Technology (2025)

work page 2025

[12] [12]

YOLOv11: An Overview of the Key Architectural Enhancements

Khanam, R., Hussain, M.: Yolov11: An overview of the key architectural enhance- ments. arXiv preprint arXiv:2410.17725 (2024)

work page internal anchor Pith review Pith/arXiv arXiv 2024

[13] [13]

Kuo, J., Wang, X., Li, X., Xu, T.: Federated learning in autonomous ve- hicles using cross-border training (2023), https://developer.nvidia.com/blog/ federated-learning-in-autonomous-vehicles-using-cross-border-training/

work page 2023

[14] [14]

IEEE Transactions on Machine Learning in Communications and Networking1, 210–224 (2023)

Lan, G., Liu, X.Y., Zhang, Y., Wang, X.: Communication-efficient federated learn- ing for resource-constrained edge devices. IEEE Transactions on Machine Learning in Communications and Networking1, 210–224 (2023)

work page 2023

[15] [15]

In: Proceedings of Neural Information Processing Systems, FLDPC Workshop (2019)

Li, D., Wang, J.: Fedmd: Heterogenous federated learning via model distillation. In: Proceedings of Neural Information Processing Systems, FLDPC Workshop (2019)

work page 2019

[16] [16]

In: 2024 International Joint Confer- ence on Neural Networks (IJCNN)

Liang, J., Li, J., Zhang, J., Zang, T.: Federated learning with data-free distillation for heterogeneity-aware autonomous driving. In: 2024 International Joint Confer- ence on Neural Networks (IJCNN). pp. 1–7. IEEE (2024)

work page 2024

[17] [17]

Computer Networks (2025)

Liao, G., Yang, Y., Feng, Z.: Personalized federated learning through self- knowledge distillation in vehicular edge computing. Computer Networks (2025)

work page 2025

[18] [18]

In: Proceedings of the European conference on computer vision (ECCV)

Ma, N., Zhang, X., Zheng, H.T., Sun, J.: Shufflenet v2: Practical guidelines for efficient cnn architecture design. In: Proceedings of the European conference on computer vision (ECCV). pp. 116–131 (2018) FedADAS 15

work page 2018

[19] [19]

Sensors 23(21), 8741 (2023)

Majeed, F., Shafique, U., Safran, M., Alfarhood, S., Ashraf, I.: Detection of drowsi- ness among drivers using novel deep convolutional neural network model. Sensors 23(21), 8741 (2023)

work page 2023

[20] [20]

In: Artificial intelligence and statistics

McMahan, B., Moore, E., Ramage, D., Hampson, S., y Arcas, B.A.: Communication-efficient learning of deep networks from decentralized data. In: Artificial intelligence and statistics. pp. 1273–1282. PMLR (2017)

work page 2017

[21] [21]

In: Proceedings of the IEEE International Conference on Image Processing (2026)

Mujtaba, A., Radchenko, G., Masana, M., Prodan, R.: Yawdd+: Frame-level an- notations for accurate yawn prediction. In: Proceedings of the IEEE International Conference on Image Processing (2026)

work page 2026

[22] [22]

In: 2025 3rd Interna- tional Conference on Federated Learning Technologies and Applications (FLTA)

Mujtaba, A., Radchenko, G., Prodan, R., Masana, M.: Federated distillation on edge devices: Efficient client-side filtering for non-iid data. In: 2025 3rd Interna- tional Conference on Federated Learning Technologies and Applications (FLTA). pp. 228–235 (2025)

work page 2025

[23] [23]

International Journal of Intelligent Systems2025(1), 7406934 (2025)

Qin, L., Zhu, T., Zhou, W., Yu, P.S.: Knowledge distillation in federated learning: A survey on long lasting challenges and new solutions. International Journal of Intelligent Systems2025(1), 7406934 (2025)

work page 2025

[24] [24]

Riya, F.F., Hoque, S., Zhao, X., Sun, J.S.: Smart driver monitoring robotic system to enhance road safety : A comprehensive review (2024)

work page 2024

[25] [25]

IEEE Internet of Things Journal10(13), 11643–11654 (2023)

Shang, E., Liu, H., Yang, Z., Du, J., Ge, Y.: Fedbikd: Federated bidirectional knowledge distillation for distracted driving detection. IEEE Internet of Things Journal10(13), 11643–11654 (2023)

work page 2023

[26] [26]

Nature Communications15(1), 349 (2024)

Shao, J., Wu, F., Zhang, J.: Selective knowledge sharing for privacy-preserving federated distillation without a good teacher. Nature Communications15(1), 349 (2024)

work page 2024

[27] [27]

In: 2024 International conference on advances in data engineering and intelligent computing systems (ADICS)

Varghese, R., Sambath, M.: Yolov8: A novel object detection algorithm with en- hanced performance and robustness. In: 2024 International conference on advances in data engineering and intelligent computing systems (ADICS). pp. 1–6. IEEE (2024)

work page 2024

[28] [28]

In: 2024 IEEE Interna- tional Conference on Big Data (BigData)

Wang, J., Gao, J.: Federated learning with knowledge distillation to mitigate catas- trophic forgetting and data heterogeneity in iov systems. In: 2024 IEEE Interna- tional Conference on Big Data (BigData). pp. 2914–2923. IEEE (2024)

work page 2024

[29] [29]

Wang, X., Tang, Z., Guo, J., Meng, T., Wang, C., Wang, T., Jia, W.: Empowering edgeintelligence:Acomprehensivesurveyonon-deviceaimodels.ACMComputing Surveys57(9), 1–39 (2025)

work page 2025

[30] [30]

IEEE Transac- tions on Vehicular Technology (2025)

Xiao, S., Huang, X., Zhou, M., Liang, C., Chen, Q.: Feddld: Dual-level federated distillation with adaptive knowledge transfer for dag-secured iovs. IEEE Transac- tions on Vehicular Technology (2025)

work page 2025

[31] [31]

IEICE Transactions on Information and Systems p

XU, M., ZHAN, A., WU, C., WANG, Z.: A novel driver fatigue detection method based on dual-stream swin-transformer. IEICE Transactions on Information and Systems p. 2024EDL8094 (2025)

work page 2025

[32] [32]

In: International Conference on Learning Representations (2023)

Zhang, J., Chen, C., Lyu, L.: Ideal: Query-efficient data-free learning from black- box models. In: International Conference on Learning Representations (2023)

work page 2023

[33] [33]

In: Proc

Zhang, Y., Jia, R., Pei, H., Wang, W., Li, B., Song, D.: The secret revealer: Generative model-inversion attacks against deep neural networks. In: Proc. of the IEEE/CVF conference on computer vision and pattern recognition. pp. 253–261 (2020)

work page 2020

[34] [34]

In: 2021 33rd Chinese Control and Decision Confer- ence (CCDC)

Zhou, C., Li, J.: A real-time driver fatigue monitoring system based on lightweight convolutional neural network. In: 2021 33rd Chinese Control and Decision Confer- ence (CCDC). pp. 1548–1553. IEEE (2021)

work page 2021