arxiv: 2604.12183 · v1 · submitted 2026-04-14 · 💻 cs.LG · cs.AI· cs.CR

Recognition: unknown

Clustering-Enhanced Domain Adaptation for Cross-Domain Intrusion Detection in Industrial Control Systems

Luyao Wang

Authors on Pith no claims yet

Pith reviewed 2026-05-10 15:50 UTC · model grok-4.3

classification 💻 cs.LG cs.AIcs.CR

keywords domain adaptationintrusion detectionindustrial control systemsclusteringtransfer learningunknown attacksfeature alignmentK-Medoids

0 comments

The pith

A clustering-enhanced domain adaptation method improves unknown attack detection in industrial control systems by up to 49% over baselines.

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

The paper establishes that aligning source and target industrial control system traffic into a shared latent space through spectral-transform feature alignment, combined with K-Medoids clustering and PCA dimensionality reduction, enables effective cross-domain intrusion detection despite limited labels and shifting distributions. A sympathetic reader would care because ICS environments frequently change their traffic patterns and face new attacks, causing conventional detectors to fail without expensive new labeling. The method iteratively reduces distribution discrepancies while the clustering step improves correlation estimates and avoids heavy manual tuning. Experiments confirm substantial gains in accuracy, F-score, and stability compared to five baseline models, with the clustering component adding further boosts.

Core claim

The clustering-enhanced domain adaptation framework projects source and target ICS traffic domains into a shared latent subspace through spectral-transform-based feature alignment to iteratively reduce distribution discrepancies, while a K-Medoids clustering strategy combined with PCA-based dimensionality reduction improves cross-domain correlation estimation and reduces degradation from manual parameter tuning, leading to significantly better detection of unknown attacks.

What carries the argument

Clustering-enhanced domain adaptation framework, consisting of spectral-transform feature alignment for shared subspace projection and K-Medoids-PCA clustering for correlation enhancement.

If this is right

Unknown attack detection accuracy rises by up to 49% compared with five baseline models.
F-score gains are larger and stability is stronger across tasks.
The clustering enhancement strategy alone increases detection accuracy by up to 26% on representative tasks.
The approach reduces the impact of data scarcity and domain shift in dynamic industrial environments.

Where Pith is reading between the lines

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

If the shared-structure premise holds more broadly, the method could lower the cost of labeling data for each new ICS deployment.
The reduced need for manual tuning via clustering and PCA might simplify integration into real-time ICS monitoring pipelines.
The stability observed suggests the framework could be tested on other network security domains that experience traffic shifts, such as IoT or enterprise networks.

Load-bearing premise

Source and target ICS traffic domains share enough structural similarity that spectral alignment produces a useful common space and K-Medoids with PCA reliably improves correlation estimates without introducing bias or harming performance on the tested distributions.

What would settle it

Applying the method to new ICS traffic datasets with substantially different distributions or attack profiles and observing no accuracy gains over the baselines or no additional boost from the clustering step would falsify the central claim.

Figures

Figures reproduced from arXiv: 2604.12183 by Luyao Wang.

**Figure 2.** Figure 2: Overview of the proposed clustering-enhanced domain adaptation framework for cross-domain intrusion [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Figure 3: Illustration of the clustering enhancement module. After PCA-based feature homogenization, source-domain and target-domain samples are partitioned into clusters by K-Medoids. Cluster-level similarity is then computed to establish source–target structural correspondence, which guides the subsequent latent-space alignment. Algorithm 1 Clustering-Enhanced Domain Adaptation for ICS Intrusion Detection Input: S… view at source ↗

**Figure 4.** Figure 4: Performance comparison of the proposed method and baseline models on cross-domain unknown attack detection tasks. The results show that the proposed framework consistently achieves superior accuracy and F-score under different source–target transfer settings. Another noteworthy observation from [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗

**Figure 5.** Figure 5: Visual comparison of the proposed framework and baseline models on four cross-domain unknown attack detection tasks. The upper panel reports the accuracy results, while the lower panel presents the F-score results. The proposed method consistently achieves the best performance across different transfer settings, demonstrating its superiority in both classification correctness and class-balanced detection e… view at source ↗

read the original abstract

Industrial control systems operate in dynamic environments where traffic distributions vary across scenarios, labeled samples are limited, and unknown attacks frequently emerge, posing significant challenges to cross-domain intrusion detection. To address this issue, this paper proposes a clustering-enhanced domain adaptation method for industrial control traffic. The framework contains two key components. First, a feature-based transfer learning module projects source and target domains into a shared latent subspace through spectral-transform-based feature alignment and iteratively reduces distribution discrepancies, enabling accurate cross-domain detection. Second, a clustering enhancement strategy combines K-Medoids clustering with PCA-based dimensionality reduction to improve cross-domain correlation estimation and reduce performance degradation caused by manual parameter tuning. Experimental results show that the proposed method significantly improves unknown attack detection. Compared with five baseline models, it increases detection accuracy by up to 49%, achieves larger gains in F-score, and demonstrates stronger stability. Moreover, the clustering enhancement strategy further boosts detection accuracy by up to 26% on representative tasks. These results suggest that the proposed method effectively alleviates data scarcity and domain shift, providing a practical solution for robust cross-domain intrusion detection in dynamic industrial environments.

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

The paper adds K-Medoids clustering and PCA to domain adaptation for ICS intrusion detection and reports large accuracy gains, but those gains sit on single-point estimates without variance or tests.

read the letter

The core contribution is a two-part pipeline for cross-domain ICS intrusion detection: spectral-transform alignment to build a shared feature space across source and target traffic, plus a K-Medoids-plus-PCA clustering step meant to tighten correlation estimates and cut sensitivity to manual parameters. The abstract positions this as a direct response to scarce labels and shifting distributions in industrial settings, and the claimed lifts—up to 49% accuracy over five baselines, bigger F-score gains, and an extra 26% from the clustering piece—are the numbers a practitioner would notice first. If the deltas survive proper checking, the method could be a useful off-the-shelf option for environments that cannot relabel every new deployment. The description of the alignment loop and the clustering enhancement is clear enough that someone could reimplement the skeleton without much guesswork. The problem framing is also honest about the practical constraints of ICS traffic. The main weakness is the evaluation. No dataset details, no baseline specifications, no error bars, no seed counts, and no significance tests appear in the provided text, so the 49% and 26% figures remain single-run point estimates that could shift with different partitions or initializations. The assumption that the domains share enough structure for the alignment to help is stated but not probed with ablations or sensitivity checks. That leaves open the possibility that the clustering step sometimes helps and sometimes adds noise. The work is aimed at applied researchers and engineers who already work on intrusion detection for critical infrastructure and need something that handles domain shift without new labels. A reader who wants a concrete starting pipeline could extract value from the high-level design, though they would still need to run their own controls. It is worth sending to peer review because the setting is relevant and the idea is simple to test; referees would mainly need to insist on the missing robustness numbers and dataset transparency before any stronger claims could be accepted.

Referee Report

3 major / 1 minor

Summary. The paper proposes a clustering-enhanced domain adaptation method for cross-domain intrusion detection in industrial control systems. The framework includes a feature-based transfer learning module that uses spectral-transform-based feature alignment to project source and target domains into a shared latent subspace while iteratively reducing distribution discrepancies, plus a clustering enhancement strategy that applies K-Medoids clustering combined with PCA-based dimensionality reduction to improve cross-domain correlation estimates. The authors report that the method significantly improves unknown attack detection, achieving up to 49% higher detection accuracy than five baseline models, larger gains in F-score, stronger stability, and an additional up to 26% accuracy boost from the clustering strategy on representative tasks.

Significance. If the empirical performance gains can be substantiated with proper statistical validation, dataset details, and controls, the work could offer a practical approach to mitigating data scarcity and domain shift in ICS intrusion detection, addressing a relevant challenge in securing dynamic industrial environments where unknown attacks emerge frequently.

major comments (3)

[Abstract and Experimental Results] Abstract and Experimental Results section: The central claims of up to 49% accuracy improvement over baselines and a 26% boost from the clustering enhancement are presented as single-point estimates without dataset descriptions, baseline implementation details, error bars, number of random seeds, statistical significance tests, or ablation controls, preventing evaluation of whether the deltas exceed run-to-run variation or arise from post-hoc selection on the specific ICS traffic splits.
[Method] Method section: The spectral-transform-based feature alignment and iterative discrepancy reduction are described at a high level without explicit equations defining the shared latent subspace construction or the alignment objective, making it difficult to assess the method's grounding or whether the reported gains reduce to quantities defined by the fitted parameters (e.g., number of clusters or PCA dimensionality).
[Experimental Results] Experimental Results section: The stability and correlation improvement claims for K-Medoids+PCA rest on the assumption that source and target ICS domains share sufficient structure for the alignment to produce useful subspaces without introducing selection bias, yet no evidence (such as sensitivity analysis or cross-validation across partitions) is provided to support this for the tested traffic distributions.

minor comments (1)

[Abstract] The abstract refers to 'five baseline models' without naming them; these should be explicitly listed with references or implementation details in the main text for reproducibility.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments, which highlight important areas for strengthening the manuscript. We address each major comment below and commit to revisions that will incorporate the requested details, equations, and analyses.

read point-by-point responses

Referee: [Abstract and Experimental Results] Abstract and Experimental Results section: The central claims of up to 49% accuracy improvement over baselines and a 26% boost from the clustering enhancement are presented as single-point estimates without dataset descriptions, baseline implementation details, error bars, number of random seeds, statistical significance tests, or ablation controls, preventing evaluation of whether the deltas exceed run-to-run variation or arise from post-hoc selection on the specific ICS traffic splits.

Authors: We agree that single-point estimates limit the interpretability of the reported gains. In the revised manuscript, we will expand the Experimental Results section with: full dataset descriptions (including ICS traffic characteristics, sizes, and how source/target splits were formed); implementation details and hyperparameters for all five baselines; performance as means and standard deviations over at least five random seeds; statistical significance tests (e.g., paired t-tests or Wilcoxon tests) against baselines; and ablation studies that isolate the clustering component. These additions will demonstrate that the up-to-49% accuracy and up-to-26% clustering boosts exceed run-to-run variation and are not artifacts of particular splits. revision: yes
Referee: [Method] Method section: The spectral-transform-based feature alignment and iterative discrepancy reduction are described at a high level without explicit equations defining the shared latent subspace construction or the alignment objective, making it difficult to assess the method's grounding or whether the reported gains reduce to quantities defined by the fitted parameters (e.g., number of clusters or PCA dimensionality).

Authors: The referee correctly identifies that the current description is insufficiently formal. We will revise the Method section to include explicit mathematical formulations: the spectral transform operator, the projection into the shared latent subspace, the alignment objective (including the discrepancy measure and iterative update rule), and the precise integration of K-Medoids clustering with PCA dimensionality reduction. We will also state how the number of clusters and PCA dimensions are selected and their role in the objective, enabling readers to evaluate the method's grounding and reproducibility. revision: yes
Referee: [Experimental Results] Experimental Results section: The stability and correlation improvement claims for K-Medoids+PCA rest on the assumption that source and target ICS domains share sufficient structure for the alignment to produce useful subspaces without introducing selection bias, yet no evidence (such as sensitivity analysis or cross-validation across partitions) is provided to support this for the tested traffic distributions.

Authors: We acknowledge that additional empirical validation is required to support the assumptions underlying the clustering enhancement. In the revision, we will add a sensitivity analysis that varies the number of K-Medoids clusters and the retained PCA dimensionality, reporting detection accuracy and stability metrics across these choices. We will also include cross-validation results over multiple random partitions of the source and target traffic data to show that performance gains remain consistent and that no selection bias is introduced by the particular splits used in the original experiments. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical method with external validation

full rationale

The paper presents an algorithmic framework for cross-domain intrusion detection consisting of spectral-transform feature alignment followed by K-Medoids+PCA clustering. These steps are described procedurally and evaluated via direct comparison against five baselines on held-out ICS traffic tasks, with reported accuracy and F-score deltas. No equations, closed-form derivations, or parameter-fitting procedures are shown that would make the claimed performance gains (49% accuracy, 26% clustering boost) equivalent to the method's own inputs by construction. No self-citations are invoked as load-bearing uniqueness theorems or ansatzes. The result is therefore a standard empirical proposal whose validity rests on external test-set measurements rather than internal definitional equivalence.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The approach rests on standard machine-learning assumptions about domain alignment and clustering utility rather than new physical entities; free parameters such as cluster count and alignment iteration count are implicit but not quantified in the abstract.

free parameters (2)

Number of clusters for K-Medoids
Selected to improve cross-domain correlation estimation; exact value and selection procedure not stated in abstract.
Dimensionality after PCA
Chosen to reduce noise before clustering; specific retained dimensions not reported.

axioms (2)

domain assumption Source and target ICS traffic distributions can be aligned in a shared latent subspace via spectral-transform feature mapping
Invoked to enable iterative discrepancy reduction between domains.
domain assumption K-Medoids clustering after PCA yields more reliable cross-domain correlation estimates than manual parameter tuning alone
Central justification for the clustering enhancement component.

pith-pipeline@v0.9.0 · 5491 in / 1498 out tokens · 36364 ms · 2026-05-10T15:50:54.096532+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

41 extracted references · 6 canonical work pages

[1]

Preprocess source and target data
[2]

Apply PCA to obtainZ𝑠 andZ 𝑡
[3]

Perform K-Medoids clustering on both domains
[4]

Compute cluster-level similarity and build correspondence setΩ
[5]

Learn latent projection𝐴by minimizing the adaptation objective
[6]

Train classifier𝑔on adapted source features
[7]

After PCA transformation, the source and target samples are partitioned into𝐾 𝑠 and𝐾 𝑡 clusters, respectively: C𝑠 ={𝐶 𝑠 1 , 𝐶𝑠 2 ,

Predict target labelsˆ𝑌𝑡 =𝑔(𝐴 ⊤𝑍𝑡 ) integrated into a unified cross-domain intrusion detection pipeline. After PCA transformation, the source and target samples are partitioned into𝐾 𝑠 and𝐾 𝑡 clusters, respectively: C𝑠 ={𝐶 𝑠 1 , 𝐶𝑠 2 , . . . , 𝐶𝑠 𝐾𝑠 },C 𝑡 ={𝐶 𝑡 1, 𝐶𝑡 2, . . . , 𝐶𝑡 𝐾𝑡 }. Each cluster is represented by its medoid [9]: 𝑚𝑠 𝑝 ∈𝐶 𝑠 𝑝, 𝑚 𝑡 𝑞 ∈𝐶 ...
[8]

A theory of learning from different domains.Machine Learning, 79(1–2):151–175, 2010

Shai Ben-David, John Blitzer, Koby Crammer, and Fernando Pereira. A theory of learning from different domains.Machine Learning, 79(1–2):151–175, 2010

2010
[9]

Organ-agents: Virtual human physiology simulator via llms

Rihao Chang, Hongbo Jiao, Weizhi Nie, Huijie Guo,KaiXie,ZihanWu,LinZhao,YutongBai, Yongtao Ma, Lijuan Wang, et al. Organ-agents: Virtual human physiology simulator via llms. arXiv preprint arXiv:2508.14357, 2025

work page arXiv 2025
[10]

3d shape knowledge graph for cross-domain 3d shape retrieval

Rihao Chang, Yongtao Ma, Tong Hao, Weijie Wang, and Weizhi Nie. 3d shape knowledge graph for cross-domain 3d shape retrieval. CAAI Transactions on Intelligence Technology, 9(5):1199–1216, 2024

2024
[11]

Y. Chen, S. Su, D. Yu, H. He, X. Wang, Y. Ma, and H. Guo. Cross-domain industrial intrusion detection deep model trained with imbalanced data.IEEE Internet of Things Journal, 10:584–596, 2023

2023
[12]

Domain-adversarial training of neural networks.Journal of Machine Learning Research, 17(59):1–35, 2016

Yaroslav Ganin, Evgeniya Ustinova, Hana Ajakan, Pascal Germain, Hugo Larochelle, François Laviolette, Mario Marchand, and Victor Lempitsky. Domain-adversarial training of neural networks.Journal of Machine Learning Research, 17(59):1–35, 2016

2016
[13]

Point-pc: Point cloud completion guided by prior knowledge via causal inference.CAAI Transactions on Intelligence Technology, 2025

Xuesong Gao, Chuanqi Jiao, Ruidong Chen, Weijie Wang, and Weizhi Nie. Point-pc: Point cloud completion guided by prior knowledge via causal inference.CAAI Transactions on Intelligence Technology, 2025

2025
[14]

M. R. Gauthama Raman, Chuadhry Mujeeb Ahmed, and Aditya Mathur. Machine learning for intrusion detection in industrial control systems: Challenges and lessons from experimental evaluation.Cybersecurity, 4(1):27, 2021

2021
[15]

Jolliffe.Principal Component Analysis

Ian T. Jolliffe.Principal Component Analysis. Springer, New York, 2 edition, 2002

2002
[16]

Rousseeuw

Leonard Kaufman and Peter J. Rousseeuw. Partitioning around medoids (program pam). In Finding Groups in Data: An Introduction to Cluster Analysis, pages 68–125. John Wiley & Sons, New York, 1990

1990
[17]

Deep transfer learning for intrusion detection in industrial control networks: A comprehensive review.Journal of Network and Computer Applications, 220:103760, 2023

Hamza Kheddar, Yassine Himeur, and Ali Ismail Awad. Deep transfer learning for intrusion detection in industrial control networks: A comprehensive review.Journal of Network and Computer Applications, 220:103760, 2023

2023
[18]

Detecting cyberattacks in industrial control systems using convolutional neural networks.arXiv preprint arXiv:1806.08110, 2018

Moshe Kravchik and Asaf Shabtai. Detecting cyberattacks in industrial control systems using convolutional neural networks.arXiv preprint arXiv:1806.08110, 2018

work page arXiv 2018
[19]

Freeinsert: Disentangled text-guided object insertion in 3d gaussian scene without spatial priors, 2025

Chenxi Li, Weijie Wang, Qiang Li, Bruno Lepri, Nicu Sebe, and Weizhi Nie. Freeinsert: Disentangled text-guided object insertion in 3d gaussian scene without spatial priors, 2025

2025
[20]

Multimodal information fusion based on lstm for 3d model retrieval.Multimedia Tools and Applications, 79(45–46):33943–33956, 2020

Qi Liang, Ning Xu, Weijie Wang, and Xingjian Long. Multimodal information fusion based on lstm for 3d model retrieval.Multimedia Tools and Applications, 79(45–46):33943–33956, 2020

2020
[21]

Large language models for multimodal deformable image registration.arXiv preprint arXiv:2408.10703, 2024

Mingrui Ma, Weijie Wang, Jie Ning, Jianfeng He, Nicu Sebe, and Bruno Lepri. Large language models for multimodal deformable image registration.arXiv preprint arXiv:2408.10703, 2024

work page arXiv 2024
[22]

Unsupervised deep probabilistic approach for partial point cloud registration

Guofeng Mei, Hao Tang, Xiaoshui Huang, Weijie Wang, Juan Liu, Jian Zhang, Luc Van Gool, and Qiang Wu. Unsupervised deep probabilistic approach for partial point cloud registration. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023

2023
[23]

Haroon Mushtaq, S. U. Khan, M. A. Jan, A. Ullah, and H. A. Khattak. A parallel architectureforthepartitioningaroundmedoids (pam) algorithm.Sensors, 18(12):4129, 2018

2018
[24]

T2td: Text-3d generation model based on prior knowledge guidance.IEEE Transactions on Pattern Analysis and Machine Intelligence, 47(1):172–189, 2025

Weizhi Nie, Ruidong Chen, Weijie Wang, Bruno Lepri, and Nicu Sebe. T2td: Text-3d generation model based on prior knowledge guidance.IEEE Transactions on Pattern Analysis and Machine Intelligence, 47(1):172–189, 2025

2025
[25]

Characteristic views extraction modal based-on deep reinforcement learning for 3d model retrieval

Weizhi Nie, Weijie Wang, Anan Liu, and Chuang Chen. Characteristic views extraction modal based-on deep reinforcement learning for 3d model retrieval. In2019 IEEE International Conference on Image Processing (ICIP), pages 2389–2393, 2019

2019
[26]

Hgan: Holistic generative adversarial networks for two-dimensional image-based three-dimensional object retrieval

Weizhi Nie, Weijie Wang, Anan Liu, Jie Nie, and Yuxuan Su. Hgan: Holistic generative adversarial networks for two-dimensional image-based three-dimensional object retrieval. ACM Transactions on Multimedia Computing, Communications, and Applications, 15(4):1–24, 2019

2019
[27]

Tsang, James T

Sinno Jialin Pan, Ivor W. Tsang, James T. Kwok, and Qiang Yang. Domain adaptation via transfer component analysis. InProceedings of the 21st International Joint Conference on Artificial Intelligence (IJCAI), pages 1187–1192, 2009

2009
[28]

On lines and planes of closest fit to systems of points in space.The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science, 2(11):559–572, 1901

Karl Pearson. On lines and planes of closest fit to systems of points in space.The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science, 2(11):559–572, 1901

1901
[29]

Bringing masked autoencoders explicit contrastive properties for point cloud self-supervised learning

Bin Ren, Guofeng Mei, Danda Pani Paudel, Weijie Wang, Yawei Li, Mengyuan Liu, Rita Cucchiara, Luc Van Gool, and Nicu Sebe. Bringing masked autoencoders explicit contrastive properties for point cloud self-supervised learning. InProceedings of the Asian Conference on Computer Vision (ACCV), 2024

2024
[30]

A. Rigo, L. Stornaiuolo, Weijie Wang, M. Martino, Bruno Lepri, Nicu Sebe, and Weizhi Nie. Poci-diff: Position objects consistently and interactively with 3d-layout guided diffusion.arXiv preprint arXiv:2601.14056, 2026

work page arXiv 2026
[31]

MuhammadAzmiUmer,KhurumNazirJunejo, Muhammad Taha Jilani, and Aditya P. Mathur. Machine learning for intrusion detection in industrial control systems: Applications, challenges, and recommendations. International Journal of Critical Infrastructure Protection, 38:100516, 2022

2022
[32]

Mbt-polyp: A new multi-branch memory-augmented transformer for polyp segmentation.Image and Vision Computing, 163:105747, 2025

Tao Wang, Weijie Wang, Fausto Giunchiglia, Fengzhi Zhao, Ye Zhang, Duo Yu, and Guixia Liu. Mbt-polyp: A new multi-branch memory-augmented transformer for polyp segmentation.Image and Vision Computing, 163:105747, 2025

2025
[33]

U-hrmlp: Refining segmentation boundaries in histopathology images

Tao Wang, Kai Zhang, Weijie Wang, Mingrui Ma, Ye Zhang, He Zhao, and Guixia Liu. U-hrmlp: Refining segmentation boundaries in histopathology images. In2024 IEEE International Symposium on Biomedical Imaging (ISBI), pages 1–5, 2024

2024
[34]

Dynamically instance-guided adaptation: A backward-free approach for test-time domain adaptive semantic segmentation

Wei Wang, Zhun Zhong, Weijie Wang, Xi Chen, Charles Ling, Boyu Wang, and Nicu Sebe. Dynamically instance-guided adaptation: A backward-free approach for test-time domain adaptive semantic segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 24090–24099, 2023

2023
[35]

Fully-geometric cross-attention for point cloud registration

Weijie Wang, Guofeng Mei, Jian Zhang, Nicu Sebe, Bruno Lepri, and Fabio Poiesi. Fully-geometric cross-attention for point cloud registration. In3DV. IEEE, 2025

2025
[36]

Zeroreg: Zero-shot point cloudregistrationwithfoundationmodels.arXiv preprint arXiv:2312.03032, 2023

Weijie Wang, Wenqi Ren, Guofeng Mei, Bin Ren, Xiaoshui Huang, Fabio Poiesi, Nicu Sebe, and Bruno Lepri. Zeroreg: Zero-shot point cloud registration with foundation models. arXiv preprint arXiv:2312.03032, 2023

work page arXiv 2023
[37]

Uvmap-id: A controllable and personalized uv map generative model

Weijie Wang, Jichao Zhang, Chang Liu, Xia Li, Xingqian Xu, Humphrey Shi, Nicu Sebe, and Bruno Lepri. Uvmap-id: A controllable and personalized uv map generative model. InACM MM, pages 10725–10734, 2024

2024
[38]

Turn fake into real: Adversarial head turn attacks against deepfake detection

Weijie Wang, Zhengyu Zhao, Nicu Sebe, and Bruno Lepri. Turn fake into real: Adversarial head turn attacks against deepfake detection. arXiv preprint arXiv:2309.01104, 2023

work page arXiv 2023
[39]

Learning spatial-spectral dual adaptive graph embedding for multispectral and hyperspectral image fusion.Pattern Recognition, 151:110365, 2024

Xuquan Wang, Feng Zhang, Kai Zhang, Weijie Wang, Xiong Dun, and Jiande Sun. Learning spatial-spectral dual adaptive graph embedding for multispectral and hyperspectral image fusion.Pattern Recognition, 151:110365, 2024

2024
[40]

Structure causal models and llms integration in medical visual question answering.IEEE Transactions on Medical Imaging, 44(8):3476–3489, 2025

Zibo Xu, Qiang Li, Weizhi Nie, Weijie Wang, and Anan Liu. Structure causal models and llms integration in medical visual question answering.IEEE Transactions on Medical Imaging, 44(8):3476–3489, 2025

2025
[41]

Transfer learning for detecting unknown network attacks

Juan Zhao, Sachin Shetty, Jan Wei Pan, Charles Kamhoua, and Kevin Kwiat. Transfer learning for detecting unknown network attacks. EURASIP Journal on Information Security, 2019(1):1, 2019

2019