FedCMB: A Cross-Modal Federated Learning Framework for Multi-Modal Cyberbullying Detection with Shared Feature Space

Weiqi ZHANG; Chengzu DONG; Peng WANG

doi:10.1109/DSC67331.2025.00048

FedCMB: A Cross-Modal Federated Learning Framework for Multi-Modal Cyberbullying Detection with Shared Feature Space

Weiqi ZHANG
, Chengzu DONG^*
, Peng WANG

^*Corresponding author for this work

School of Data Science

Research output: Book Chapters | Papers in Conference Proceedings › Conference paper (refereed) › Referred Conference Paper › peer-review

Abstract

The increasing multimodality of cyberbullying poses significant challenges to detection tasks, making traditional single-modal approaches inadequate. Federated learning has emerged as an effective distributed learning paradigm for collaborative training while preserving user privacy, but existing methods generally overlook the heterogeneity of client models and modality discrepancies. To address these issues, we propose FedCMB, a cross-modal federated learning framework tailored specifically for multi-modal cyberbullying detection. FedCMB constructs a modality-invariant shared feature space to effectively fuse features from heterogeneous client models. Additionally, we introduce a knowledge distillation-based aggregation method to facilitate efficient knowledge transfer and integration from heterogeneous client models to a unified server-side model. Our experiments show that FedCMB enhances accuracy by 11.4% over baselines while preserving privacy.

Original language	English
Title of host publication	2025 IEEE 10th International Conference on Data Science in Cyberspace (DSC): Proceedings
Publisher	IEEE
Pages	319-326
Number of pages	8
ISBN (Electronic)	9798331579241
DOIs	https://doi.org/10.1109/DSC67331.2025.00048
Publication status	Published - Aug 2025
Event	2025 IEEE 10th International Conference on Data Science in Cyberspace - Baoding, China Duration: 15 Aug 2025 → 17 Aug 2025

Conference

Conference	2025 IEEE 10th International Conference on Data Science in Cyberspace
Abbreviated title	DSC 2025
Country/Territory	China
City	Baoding
Period	15/08/25 → 17/08/25

Funding

This work was supported by the Shenzhen Science and Technology Program (No. KJZD20240903103811016).

Keywords

federated learning
cyberbullying detection
multi-modal data
knowledge distillation
shared feature space
model heterogeneity

Access to Document

10.1109/DSC67331.2025.00048

Cite this

@inproceedings{2a3fbe9c95014732978343d64ecd3a15,

title = "FedCMB: A Cross-Modal Federated Learning Framework for Multi-Modal Cyberbullying Detection with Shared Feature Space",

abstract = "The increasing multimodality of cyberbullying poses significant challenges to detection tasks, making traditional single-modal approaches inadequate. Federated learning has emerged as an effective distributed learning paradigm for collaborative training while preserving user privacy, but existing methods generally overlook the heterogeneity of client models and modality discrepancies. To address these issues, we propose FedCMB, a cross-modal federated learning framework tailored specifically for multi-modal cyberbullying detection. FedCMB constructs a modality-invariant shared feature space to effectively fuse features from heterogeneous client models. Additionally, we introduce a knowledge distillation-based aggregation method to facilitate efficient knowledge transfer and integration from heterogeneous client models to a unified server-side model. Our experiments show that FedCMB enhances accuracy by 11.4\% over baselines while preserving privacy.",

keywords = "federated learning, cyberbullying detection, multi-modal data, knowledge distillation, shared feature space, model heterogeneity",

author = "Weiqi ZHANG and Chengzu DONG and Peng WANG",

year = "2025",

month = aug,

doi = "10.1109/DSC67331.2025.00048",

language = "English",

pages = "319--326",

booktitle = "2025 IEEE 10th International Conference on Data Science in Cyberspace (DSC): Proceedings",

publisher = "IEEE",

note = "2025 IEEE 10th International Conference on Data Science in Cyberspace, DSC 2025 ; Conference date: 15-08-2025 Through 17-08-2025",

}

ZHANG, W, DONG, C & WANG, P 2025, FedCMB: A Cross-Modal Federated Learning Framework for Multi-Modal Cyberbullying Detection with Shared Feature Space. in 2025 IEEE 10th International Conference on Data Science in Cyberspace (DSC): Proceedings. IEEE, pp. 319-326, 2025 IEEE 10th International Conference on Data Science in Cyberspace, Baoding, China, 15/08/25. https://doi.org/10.1109/DSC67331.2025.00048

FedCMB: A Cross-Modal Federated Learning Framework for Multi-Modal Cyberbullying Detection with Shared Feature Space. / ZHANG, Weiqi; DONG, Chengzu; WANG, Peng.
2025 IEEE 10th International Conference on Data Science in Cyberspace (DSC): Proceedings. IEEE, 2025. p. 319-326.

Research output: Book Chapters | Papers in Conference Proceedings › Conference paper (refereed) › Referred Conference Paper › peer-review

TY - GEN

T1 - FedCMB: A Cross-Modal Federated Learning Framework for Multi-Modal Cyberbullying Detection with Shared Feature Space

AU - ZHANG, Weiqi

AU - DONG, Chengzu

AU - WANG, Peng

PY - 2025/8

Y1 - 2025/8

N2 - The increasing multimodality of cyberbullying poses significant challenges to detection tasks, making traditional single-modal approaches inadequate. Federated learning has emerged as an effective distributed learning paradigm for collaborative training while preserving user privacy, but existing methods generally overlook the heterogeneity of client models and modality discrepancies. To address these issues, we propose FedCMB, a cross-modal federated learning framework tailored specifically for multi-modal cyberbullying detection. FedCMB constructs a modality-invariant shared feature space to effectively fuse features from heterogeneous client models. Additionally, we introduce a knowledge distillation-based aggregation method to facilitate efficient knowledge transfer and integration from heterogeneous client models to a unified server-side model. Our experiments show that FedCMB enhances accuracy by 11.4% over baselines while preserving privacy.

AB - The increasing multimodality of cyberbullying poses significant challenges to detection tasks, making traditional single-modal approaches inadequate. Federated learning has emerged as an effective distributed learning paradigm for collaborative training while preserving user privacy, but existing methods generally overlook the heterogeneity of client models and modality discrepancies. To address these issues, we propose FedCMB, a cross-modal federated learning framework tailored specifically for multi-modal cyberbullying detection. FedCMB constructs a modality-invariant shared feature space to effectively fuse features from heterogeneous client models. Additionally, we introduce a knowledge distillation-based aggregation method to facilitate efficient knowledge transfer and integration from heterogeneous client models to a unified server-side model. Our experiments show that FedCMB enhances accuracy by 11.4% over baselines while preserving privacy.

KW - federated learning

KW - cyberbullying detection

KW - multi-modal data

KW - knowledge distillation

KW - shared feature space

KW - model heterogeneity

U2 - 10.1109/DSC67331.2025.00048

DO - 10.1109/DSC67331.2025.00048

M3 - Conference paper (refereed)

SP - 319

EP - 326

BT - 2025 IEEE 10th International Conference on Data Science in Cyberspace (DSC): Proceedings

PB - IEEE

T2 - 2025 IEEE 10th International Conference on Data Science in Cyberspace

Y2 - 15 August 2025 through 17 August 2025

ER -

FedCMB: A Cross-Modal Federated Learning Framework for Multi-Modal Cyberbullying Detection with Shared Feature Space

Abstract

Conference

Funding

Keywords

Access to Document

Fingerprint

Cite this