pith. machine review for the scientific record. sign in

arxiv: 2605.12584 · v1 · submitted 2026-05-12 · 💻 cs.LG · cs.AI

Recognition: 2 theorem links

· Lean Theorem

Towards Robust Federated Multimodal Graph Learning under Modality Heterogeneity

Sirui Zhang , Haonan Wang , Xunkai Li , Zekai Chen , Shumeng Li , Hongchao Qin , Rong-Hua Li , Guoren Wang
Authors on Pith no claims yet

Pith reviewed 2026-05-14 21:51 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords federated learningmultimodal graphsmissing modalitiesmodality heterogeneityrobust aggregationgraph neural networksnon-IID data
0
0 comments X

The pith

FedMPO recovers missing modalities in federated multimodal graphs using topology context and reliability-weighted aggregation.

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

The paper addresses challenges in federated multimodal graph learning where data is distributed across clients and modalities are often missing. It identifies that standard approaches fail because local completion cannot use global semantics and aggregation does not account for varying reliability. To solve this, it introduces FedMPO with three components: topology-aware generation to recover features from graph context, expert routing to filter noise, and reliability-aware aggregation to down-weight bad updates. Experiments show gains in high-missing and non-IID cases, making federated learning viable for real-world incomplete multimodal graphs.

Core claim

FedMPO is a federated framework that uses topology-aware cross-modal generation to recover missing features using comprehensive graph context, missing-aware expert routing to locally filter out noisy recovered signals, and reliability-aware aggregation to appropriately down-weight unreliable updates, leading to improved performance on multimodal graph tasks under modality heterogeneity.

What carries the argument

topology-aware cross-modal generation combined with missing-aware expert routing and reliability-aware aggregation

If this is right

  • Performance improves by up to 4.10% in high-missing-rate scenarios across multiple tasks.
  • Gains reach 5.65% in non-IID federated settings compared to existing baselines.
  • The separation of client-side completion from server-side aggregation supports privacy-preserving collaboration.
  • The method applies directly to real-world network applications with isolated multimodal graphs.
  • Down-weighting unreliable client updates stabilizes training under heterogeneous modality availability.

Where Pith is reading between the lines

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

  • The reliability-aware weighting could reduce reliance on client selection heuristics in other federated setups with noisy local models.
  • Topology-aware recovery might combine with existing graph imputation methods to handle even higher missing rates.
  • Scaling tests on larger graphs could show whether the local context assumption holds beyond the evaluated datasets.
  • Similar pipelines may apply to non-graph multimodal federated tasks like image-text data in distributed settings.

Load-bearing premise

Client-side topology-aware generation can reliably recover missing modalities from local graph context alone, and the reliability metric accurately reflects true update quality without introducing new selection bias.

What would settle it

An experiment on a graph dataset with extreme modality missingness where the generated features degrade global model accuracy despite the reliability weighting, or where the reliability scores fail to correlate with actual contribution to final performance.

Figures

Figures reproduced from arXiv: 2605.12584 by Guoren Wang, Haonan Wang, Hongchao Qin, Rong-Hua Li, Shumeng Li, Sirui Zhang, Xunkai Li, Zekai Chen.

Figure 1
Figure 1. Figure 1: Overview of missing-modality learning paradigms and the proposed FEDMPO. Wan et al. [2026], Wang et al. [2023a], Wu et al. [2024b]. It uses graph context as evidence for node￾level modality generation, but assumes centralized data access and cannot benefit from cross-client knowledge. In contrast, federated missing-modality learning methods aggregate locally updated modality generators on the server and br… view at source ↗
Figure 2
Figure 2. Figure 2: The overview of the proposed FEDMPO. Missing-modality learning. Incomplete multimodal learning handles partially missing modalities through cross-modal reconstruction, generative imputation, shared–specific representation learn￾ing, and missing-aware fusion Ma et al. [2021, 2022], Wang et al. [2023a], Reza et al. [2024], Wu et al. [2024b]. Although effective in centralized or non-graph settings, these meth… view at source ↗
Figure 3
Figure 3. Figure 3: Ablation study of FEDMPO. We compare the full model with variants removing AGMG, MoE fusion, cross-modal alignment, and reliability-aware aggregation across three downstream tasks. The performance drops verify the effectiveness of each component. semantic consistency among modalities. Finally, w/o RelAgg replaces reliability-aware aggregation with standard FedAvg. Its performance drop shows that data-size-… view at source ↗
Figure 4
Figure 4. Figure 4: Sensitivity and robustness analysis of FEDMPO. (a) Node classification ACC on Ele￾Fashion and Grocery under different combinations of reconstruction coefficient λrec and routing coefficient λroute. (b) Robustness of FEDMPO under different Dirichlet α values and modality missing rates η. 4.4 Efficiency Study FEDMPO introduces topology-aware modality generation and missing-aware expert routing, but the overh… view at source ↗
Figure 5
Figure 5. Figure 5: Ablation study of FEDMPO. (a) Consistent performance drops in variants (w/o AGMG, MoE, Align) prove the necessity of each structural design. (b) Comparisons w/ and w/o RelAgg confirm that reliability-aware aggregation is critical for stabilizing performance and reducing training costs. representations. At early communication rounds, recovered features are less reliable, and the router assigns relatively hi… view at source ↗
Figure 6
Figure 6. Figure 6: System efficiency comparison across graph federated learning baselines. FedMPO achieves [PITH_FULL_IMAGE:figures/full_fig_p024_6.png] view at source ↗
read the original abstract

Recently, multimodal graph learning (MGL) has garnered significant attention for integrating diverse modality information and structured context to support various network applications. However, real-world graphs are often isolated due to data-sharing limitations across multiple parties, and their modalities are frequently incomplete. This highlights an urgent need to develop a robust federated approach. However, we find that existing methods remain insufficient. On the one hand, centralized MGL methods that handle missing modalities overlook the knowledge sharing and generalization in federated scenarios. On the other hand, while federated MGL methods have become increasingly mature, they primarily target non-graph data. Based on these technologies, we identify a two-stage pipeline wherein client-side completion reconstructs missing modalities, and server-side aggregation integrates the client-updated parameters of both the modality generator and the backbone models. Although this serves as a general solution, we identify two primary challenges in achieving greater robustness: (1) Topology-Isolated Local Completion: Client-side modality generation struggles to effectively leverage global semantics. (2) Reliability-Imbalanced Global Aggregation: Server-side multi-party collaboration is hindered by client updates with varying modality availability and recovery reliability. To address these challenges, we propose \textsc{FedMPO}, which utilizes topology-aware cross-modal generation to recover missing features using comprehensive graph context, missing-aware expert routing to locally filter out noisy recovered signals, and reliability-aware aggregation to appropriately down-weight unreliable updates. Extensive experiments on 3 tasks across 6 datasets demonstrate that FedMPO outperforms baselines, achieving performance gains of up to 4.10% and 5.65% in high-missing and non-IID settings.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces FedMPO, a federated framework for multimodal graph learning that handles missing modalities and non-IID client distributions. It identifies two challenges—Topology-Isolated Local Completion and Reliability-Imbalanced Global Aggregation—and proposes topology-aware cross-modal generation, missing-aware expert routing, and reliability-aware aggregation as solutions. Experiments across 3 tasks and 6 datasets report performance gains of up to 4.10% in high-missing-rate settings and 5.65% in non-IID settings over baselines.

Significance. If the empirical gains hold under rigorous verification and the method respects strict federated isolation, the work would advance practical federated multimodal graph learning by explicitly targeting modality incompleteness. The two-stage pipeline framing and component-wise design offer a reusable template for distributed graph tasks in privacy-sensitive domains.

major comments (2)
  1. [Abstract] Abstract and §3 (Method): The topology-aware cross-modal generation is claimed to recover missing features 'using comprehensive graph context' to solve the Topology-Isolated Local Completion challenge. In a federated setting clients hold only local graphs; the manuscript must explicitly show how global semantics are obtained (e.g., via server prototypes or shared embeddings) without data exchange. This assumption is load-bearing for the 4.10% high-missing gain, as purely local context would leave challenge (1) unaddressed.
  2. [§4] §4 (Experiments): The headline gains (4.10% and 5.65%) are reported without tabulated baseline details, statistical significance tests (e.g., paired t-tests or Wilcoxon), ablation results isolating each of the three components, or exact missing-rate and non-IID simulation protocols. These omissions prevent verification that improvements arise from the proposed mechanisms rather than dataset artifacts or implementation choices.
minor comments (2)
  1. [Abstract] Abstract: The phrase '3 tasks across 6 datasets' should name the tasks (e.g., node classification, link prediction) and datasets for immediate clarity.
  2. Notation: Define all acronyms (MGL, FedMPO, etc.) at first use and ensure consistent use of symbols for missing-rate and reliability metrics across equations and text.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We sincerely thank the referee for the constructive and detailed feedback. We address each major comment point-by-point below, providing clarifications and committing to revisions that strengthen the manuscript without altering its core claims.

read point-by-point responses

  1. Referee: [Abstract] Abstract and §3 (Method): The topology-aware cross-modal generation is claimed to recover missing features 'using comprehensive graph context' to solve the Topology-Isolated Local Completion challenge. In a federated setting clients hold only local graphs; the manuscript must explicitly show how global semantics are obtained (e.g., via server prototypes or shared embeddings) without data exchange. This assumption is load-bearing for the 4.10% high-missing gain, as purely local context would leave challenge (1) unaddressed.

    Authors: We thank the referee for this important observation. In FedMPO the topology-aware cross-modal generator is a shared model whose parameters are aggregated at the server after each round of local training. Because the generator is updated across all clients, its weights encode global cross-modal and topological patterns observed in the federated population; each client then applies the latest global generator to its own local graph. No raw features or edges are exchanged—only model parameters—thereby preserving strict federated isolation while still supplying global semantic context. We will add an explicit paragraph in the revised §3 that diagrams this flow and cites the relevant federated-learning literature on parameter-based knowledge transfer. revision: yes

  2. Referee: [§4] §4 (Experiments): The headline gains (4.10% and 5.65%) are reported without tabulated baseline details, statistical significance tests (e.g., paired t-tests or Wilcoxon), ablation results isolating each of the three components, or exact missing-rate and non-IID simulation protocols. These omissions prevent verification that improvements arise from the proposed mechanisms rather than dataset artifacts or implementation choices.

    Authors: We fully agree that these elements are required for rigorous verification. In the revised §4 we will (i) present complete tables listing every baseline with mean and standard deviation, (ii) report paired t-test p-values (and Wilcoxon signed-rank results where appropriate) over 5 independent runs, (iii) add a dedicated ablation table that removes each of the three proposed modules in turn, and (iv) specify the exact missing-rate schedules (uniform random modality dropout at 30/50/70 %) together with the Dirichlet concentration parameters used to generate non-IID client partitions. These additions will allow readers to confirm that the reported gains originate from the proposed mechanisms. revision: yes

Circularity Check

0 steps flagged

No significant circularity; components defined on standard primitives with independent experimental validation.

full rationale

The paper identifies two challenges from limitations of prior centralized and federated MGL methods, then defines FedMPO via three explicit components (topology-aware cross-modal generation, missing-aware expert routing, reliability-aware aggregation) operating on standard graph and federated learning primitives. No equations or derivations reduce claimed performance gains to fitted parameters, self-referential quantities, or self-citation chains; the abstract and method description present algorithmic steps without tautological equivalence to inputs. Experimental results on 6 datasets are reported as external validation rather than constructed from the method definition itself. This is the expected non-circular outcome for a method-proposal paper.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The approach rests on standard domain assumptions from federated learning and graph neural networks without introducing new free parameters, axioms beyond those, or invented entities in the abstract description.

axioms (2)
  • domain assumption Clients share only model parameters and not raw data while still benefiting from global graph structure.
    Invoked in the federated setup and topology-aware generation step.
  • domain assumption Graph topology provides useful context for recovering missing modality features.
    Central to the topology-aware cross-modal generation component.

pith-pipeline@v0.9.0 · 5621 in / 1365 out tokens · 48545 ms · 2026-05-14T21:51:46.516152+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

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

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

Reference graph

Works this paper leans on

39 extracted references · 39 canonical work pages · 1 internal anchor

  1. [1]

    Proceedings of the 20th International Conference on Artificial Intelligence and Statistics (AISTATS) , series =

    McMahan, Brendan and Moore, Eider and Ramage, Daniel and Hampson, Seth and Arcas, Blaise Ag\"uera y , title =. Proceedings of the 20th International Conference on Artificial Intelligence and Statistics (AISTATS) , series =

    work page

  2. [2]

    Proceedings of Machine Learning and Systems , volume =

    Li, Tian and Sahu, Anit Kumar and Talwalkar, Ameet and Smith, Virginia , title =. Proceedings of Machine Learning and Systems , volume =

    work page

  3. [3]

    Proceedings of the 37th International Conference on Machine Learning (ICML) , series =

    Karimireddy, Sai Praneeth and Kale, Satyen and Mohri, Mehryar and Reddi, Sashank and Stich, Sebastian and Suresh, Ananda Theertha , title =. Proceedings of the 37th International Conference on Machine Learning (ICML) , series =

    work page

  4. [4]

    Machine Learning and Knowledge Discovery in Databases

    Che, Liang and Wang, Jian and Liu, Xiaoyang and Ma, Fenglong , title =. Machine Learning and Knowledge Discovery in Databases. Research Track , pages =

    work page

  5. [5]

    and Le Nguyen, Phuong and Huynh, Trung Tin , title =

    Nguyen, Minh Duc and Nguyen, Thanh Tam and Pham, Huy Hoang and Hoang, Tuan N. and Le Nguyen, Phuong and Huynh, Trung Tin , title =. 2024 22nd International Symposium on Network Computing and Applications (NCA) , pages =

    work page 2024

  6. [6]

    ICASSP 2024 - 2024 IEEE International Conference on Acoustics, Speech and Signal Processing , pages =

    Peng, Yu and Bian, Jiang and Xu, Jingren , title =. ICASSP 2024 - 2024 IEEE International Conference on Acoustics, Speech and Signal Processing , pages =

    work page 2024

  7. [7]

    Advances in Neural Information Processing Systems , volume =

    Wu, Fangzhao and Wang, Xuezhi and Wang, Yi and Liu, Tianjian and Su, Lulu and Gao, Jianfeng , title =. Advances in Neural Information Processing Systems , volume =

    work page

  8. [8]

    arXiv preprint arXiv:2405.06822 , year =

    Xie, Liang and Lin, Ming and Luan, Tuan and Li, Chao and Fang, Yixuan and Shen, Qitao and Wu, Zongwei , title =. arXiv preprint arXiv:2405.06822 , year =

    work page arXiv

  9. [9]

    Proceedings of the VLDB Endowment , volume =

    Li, Xunkai and Zhu, Yinlin and Pang, Boyang and Yan, Guochen and Yan, Yeyu and Li, Zening and Wu, Zhengyu and Zhang, Wentao and Li, Rong-Hua and Wang, Guoren , title =. Proceedings of the VLDB Endowment , volume =

    work page

  10. [10]

    and Welling, Max , title =

    Kipf, Thomas N. and Welling, Max , title =. International Conference on Learning Representations (ICLR) , year =

    work page

  11. [11]

    and Ying, Rex and Leskovec, Jure , title =

    Hamilton, William L. and Ying, Rex and Leskovec, Jure , title =. Advances in Neural Information Processing Systems , volume =

    work page

  12. [12]

    Graph Attention Networks , booktitle =

    Veli. Graph Attention Networks , booktitle =

    work page

  13. [13]

    International Conference on Learning Representations (ICLR) , year =

    Xu, Keyulu and Hu, Weihua and Leskovec, Jure and Jegelka, Stefanie , title =. International Conference on Learning Representations (ICLR) , year =

    work page

  14. [14]

    Proceedings of the 27th ACM International Conference on Multimedia , pages =

    Wei, Yinwei and Wang, Xiang and Nie, Liqiang and He, Xiangnan and Hong, Richang and Chua, Tat-Seng , title =. Proceedings of the 27th ACM International Conference on Multimedia , pages =

    work page

  15. [15]

    Information Processing & Management , volume =

    Tao, Zhiqiang and Wei, Yinwei and Wang, Xiang and He, Xiangnan and Huang, Xianglin and Chua, Tat-Seng , title =. Information Processing & Management , volume =

    work page

  16. [16]

    Neural Computing and Applications , volume =

    Jia, Xiaoxu and Jiang, Min and Dong, Yuxiao and Zhu, Fei and Lin, Huajie and Xin, Yuzhong and Chen, Hong , title =. Neural Computing and Applications , volume =

    work page

  17. [17]

    Proceedings of the ACM on the Web Conference 2025 , pages =

    He, Yufei and Sui, Yuan and He, Xiaoxin and Liu, Yue and Sun, Yifei and Hooi, Bryan , title =. Proceedings of the ACM on the Web Conference 2025 , pages =

    work page 2025

  18. [18]

    arXiv preprint arXiv:2402.05322 , year =

    Peng, Chao and He, Jia and Xia, Feng , title =. arXiv preprint arXiv:2402.05322 , year =

    work page arXiv

  19. [19]

    Nature Machine Intelligence , volume =

    Ektefaie, Yasha and Dasoulas, George and Noori, Ayush and Farhat, Maha and Zitnik, Marinka , title =. Nature Machine Intelligence , volume =

    work page

  20. [20]

    Advances in Neural Information Processing Systems , volume =

    Hu, Weihua and Fey, Matthias and Zitnik, Marinka and Dong, Yuxiao and Ren, Hongyu and Liu, Bowen and Catasta, Michele and Leskovec, Jure , title =. Advances in Neural Information Processing Systems , volume =

    work page

  21. [21]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages =

    Zhu, Jing and Zhou, Yuhang and Qian, Shengyi and He, Zhongmou and Zhao, Tong and Shah, Neil and Koutra, Danai , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages =

    work page

  22. [22]

    arXiv preprint arXiv:2410.09132 , year =

    Yan, Huikai and Li, Chengzhuo and Yu, Zihan and Yin, Jiahao and Liu, Rui and Zhang, Peiyan and Han, Wenze and Li, Meng and Zeng, Zhiyuan and Sun, Honglei and Deng, Wenrui and Sun, Fan and Zhang, Qian and Wang, Sheng , title =. arXiv preprint arXiv:2410.09132 , year =

    work page arXiv

  23. [23]

    arXiv preprint arXiv:2602.05576 , year =

    Wan, Chenxi and Li, Xunkai and Zuo, Yilong and Deng, Haokun and Li, Sihan and Fan, Bowen and Qin, Hongchao and Li, Ronghua and Wang, Guoren , title =. arXiv preprint arXiv:2602.05576 , year =

    work page arXiv

  24. [24]

    arXiv preprint arXiv:2601.22416 , year =

    Li, Xunkai and Ai, Yuming and Zhu, Yinlin and Lu, Haodong and Zhang, Yi and Fu, Guohao and Fan, Bowen and Dai, Qiangqiang and Li, Rong-Hua and Wang, Guoren , title =. arXiv preprint arXiv:2601.22416 , year =

    work page arXiv

  25. [25]

    Pattern Recognition , year =

    Wang, Wenhai and others , title =. Pattern Recognition , year =

    work page

  26. [26]

    Proceedings of the AAAI Conference on Artificial Intelligence , year =

    Zhao, Xiaoming and others , title =. Proceedings of the AAAI Conference on Artificial Intelligence , year =

    work page

  27. [27]

    arXiv preprint arXiv:2407.19108 , year =

    Wu, Jiayi and others , title =. arXiv preprint arXiv:2407.19108 , year =

    work page arXiv

  28. [28]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages =

    Girdhar, Rohit and El-Nouby, Alaaeldin and Liu, Zhuang and Singh, Mannat and Alwala, Kalyan Vasudev and Joulin, Armand and Misra, Ishan , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages =

    work page

  29. [29]

    Proceedings of the 38th International Conference on Machine Learning (ICML) , series =

    Radford, Alec and Kim, Jong Wook and Hallacy, Chris and Ramesh, Aditya and Goh, Gabriel and Agarwal, Sandhini and Sastry, Girish and Askell, Amanda and Mishkin, Pamela and Clark, Jack and Krueger, Gretchen and Sutskever, Ilya , title =. Proceedings of the 38th International Conference on Machine Learning (ICML) , series =

    work page

  30. [30]

    International Conference on Learning Representations , year =

    Dosovitskiy, Alexey and Beyer, Lucas and Kolesnikov, Alexander and Weissenborn, Dirk and Zhai, Xiaohua and Unterthiner, Thomas and Dehghani, Mostafa and Minderer, Matthias and Heigold, Georg and Gelly, Sylvain and Uszkoreit, Jakob and Houlsby, Neil , title =. International Conference on Learning Representations , year =

    work page

  31. [31]

    Oquab, Maxime and Darcet, Timoth\'ee and Moutakanni, Theo and Vo, Huy V. and Szafraniec, Marc and Khalidov, Vasil and Fernandez, Pierre and Haziza, Daniel and Massa, Francisco and El-Nouby, Alaaeldin and Assran, Mahmoud and Ballas, Nicolas and Galuba, Wojciech and Howes, Russell and Huang, Po-Yao and Li, Shang-Wen and Misra, Ishan and Rabbat, Mike and Sha...

    work page

  32. [32]

    , title =

    Raffel, Colin and Shazeer, Noam and Roberts, Adam and Lee, Katherine and Narang, Sharan and Matena, Michael and Zhou, Yanqi and Li, Wei and Liu, Peter J. , title =. Journal of Machine Learning Research , volume =

    work page

  33. [33]

    Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing (EMNLP-IJCNLP) , pages =

    Ni, Jianmo and Li, Jiacheng and McAuley, Julian , title =. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing (EMNLP-IJCNLP) , pages =

    work page 2019

  34. [34]

    Bridging Language and Items for Retrieval and Recommendation: Benchmarking LLMs as Semantic Encoders

    Hou, Yupeng and Li, Jiacheng and He, Zhankui and Yan, An and Chen, Xiusi and McAuley, Julian , title =. arXiv preprint arXiv:2403.03952 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  35. [35]

    Proceedings of the 12th ACM Conference on Recommender Systems , pages =

    Wan, Mengting and McAuley, Julian , title =. Proceedings of the 12th ACM Conference on Recommender Systems , pages =

    work page

  36. [36]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    SMIL: Multimodal Learning with Severely Missing Modality , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    work page

  37. [37]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Are Multimodal Transformers Robust to Missing Modality? , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  38. [38]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Multi-Modal Learning with Missing Modality via Shared-Specific Feature Modelling , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  39. [39]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

    Robust Multimodal Learning with Missing Modalities via Parameter-Efficient Adaptation , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

    work page