Pith. sign in

REVIEW 4 major objections 6 minor 52 references

Remember Past, Anticipate Future: Learning Continual Multimodal Misinformation Detectors

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a continual multimodal misinformation detector combining a Dirichlet-process-expanded mixture of experts with a neural-ODE model of evolving fake-article distributions consistently and significantly outperforms six…

desk verdict Load-bearing ambiguity in the dynamics training objective (Eq. 12) makes the 'anticipate future' gains unverifiable as written, but the empirical breadth and method deserve a serious referee. read the letter →

arxiv 2507.05939 v1 pith:3JIK4NAZ submitted 2025-07-08 cs.CL cs.MM

Bing Wang , Ximing Li , Mengzhe Ye , Changchun Li , Bo Fu , Jianfeng Qu , Lin Yuanbo Wu This is my paper · ORCID
classification cs.CLcs.MM
keywords continuallearningmultimodalmisinformationdetectioncatastrophicforgettingmixtureofexpertsDirichletprocessneuralordinarydifferentialequationsenvironmentaldistributiondynamicssocialmedia
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tackles a practical gap: misinformation detectors are trained offline but deployed on a stream of new events, where performance on old events drops and future events have shifted distributions. It proposes DaedCMD, a continual detector with two modules: a Dirichlet-process-expanded mixture of experts that isolates event-specific parameters to reduce forgetting, and a continuous-time dynamics model that predicts the fake-article feature distribution for future events. The paper's central claim is that this combination consistently outperforms six multimodal misinformation baselines and three continual learning methods on GossipCop, Weibo, and Twitter, with the largest gains on the smallest, most imbalanced dataset. If true, it would give practitioners a drop-in way to keep misinformation detectors current on live social media streams.

What carries the argument

The central object carrying the future-anticipation claim is the environmental distribution $P(\hat{e}|y=1)=\mathcal{N}(\mu,\sigma^2)$, a Gaussian over the multimodal features of fake articles, whose mean and covariance evolve in continuous time through the neural ordinary differential equations $\frac{d\mu}{dt}=G_{\phi_\mu}(\mu,t)$ and $\frac{d\sigma}{dt}=G_{\phi_\sigma}(\sigma,t)$, integrated with the dopri5 solver. This lets the model sample an environmental feature $\hat{e}$ from a predicted future distribution and concatenate it with the forgetting-resistant feature for veracity prediction. The forgetting-prevention machinery is the dynamically adapted mixture of experts, where a Dirichlet-process responsibility in Eq. (4) decides when to spawn a new event-specific expert and an exponential-moving-average update protects the event-shared expert from interference.

What would settle it

Train DaedCMD under a strict streaming protocol in which the integral in Eq. (12) is evaluated only over already-seen events and then test on the next future event; if the accuracy gains over LoRAMoE shrink to within noise, the reported advantage depends on access to future events' distributions during training.

Watch

Extended reading notes

Core claim

DaedCMD claims to solve continual multimodal misinformation detection by combining two mechanisms. To remember past knowledge, it uses a mixture-of-experts structure with event-shared experts and event-specific experts, where a Dirichlet-process responsibility score decides when a genuinely new event deserves its own expert rather than allocating one per event. To anticipate future knowledge, it models the environmental distribution of fake articles as a Gaussian distribution over multimodal features and learns a continuous-time dynamics model, trained with neural ordinary differential equations, that predicts the mean and covariance of that distribution at future time points. The paper reports that adding DaedCMD to six existing multimodal misinformation detectors and three continual learning baselines consistently improves accuracy and F1 across three datasets, with performance gains ordered GossipCop less than Weibo less than Twitter.

Load-bearing premise

The future-prediction objective trains the dynamics model against the true distribution of every event, including events that appear later in the stream, so the 'prediction' is not a held-out prediction unless the loss is restricted to past events.

Editorial extensions

If this is right

  • Deploying DaedCMD on a live misinformation feed should slow the accuracy decline on older events that continuing training normally causes.
  • Because DaedCMD wraps the base detector, the same continual-learning modules can be reused to upgrade existing multimodal misinformation detectors rather than retraining them from scratch.
  • Because experts are created only when a new event is sufficiently different, the parameter count grows with the number of genuinely novel events, not with every batch.
  • On small and imbalanced datasets, the environmental-dynamics feature supplies information that sparse training data cannot, so the expected practical gain is largest where detection is hardest.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper: a strict online evaluation would restrict the dynamics loss in Eq. (12) to already-seen events; the paper does not state which protocol was used, so the future-prediction gains may depend on access to later events' distributions.
  • Beyond the paper: because the dynamics model tracks only the fake class, extending it to model the real-class distribution could improve calibration and provide a testable variant.
  • Beyond the paper: replacing the clustering-based split with a pure timestamp split would test whether the gains persist on genuinely time-ordered streams rather than semantically clustered events.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper introduces DAEDCMD, a multimodal misinformation detector designed for continual data streams. It combines a mixture-of-experts architecture with a Dirichlet-process-style dynamic expansion of event-specific experts to mitigate catastrophic forgetting, and a continuous-time dynamics model based on neural ODEs to predict the evolution of a Gaussian 'environmental distribution' of fake-article features. The model is evaluated on three continual MMD datasets (GossipCop, Weibo, Twitter), applied on top of six MMD baselines and compared to three continual-learning baselines, reporting consistent accuracy/F1 gains. The paper also includes ablations, a forgetting analysis, a parameter sensitivity study, and a case study.

Significance. Continual misinformation detection is a practical and timely problem, and the paper's combination of parameter isolation with an explicit model of distribution dynamics is interesting. The empirical evaluation is broad: six MMD backbones, three datasets, and three continual-learning baselines, with a public code link and ablations that isolate the contribution of each module. If the future-anticipation mechanism is shown to be causally valid—that is, if the dynamics model is trained without access to future events' statistics—the results would be a solid step toward deployment in streaming settings. At present, that validity is not established because the training objective in Eq. (12) is ambiguous about whether future-event statistics are used.

major comments (4)
  1. [§3.3, Eq. (12); Algorithm 1] The dynamics-model objective L_DM in Eq. (12) is defined as (1/K)∫_0^K [‖μ̂_τ−μ_τ‖₂²+‖σ̂_τ−σ_τ‖₂²] dt, where μ_τ and σ_τ are the ground-truth mean and covariance of all data whose temporal label is τ. Under the sequential training loop in Algorithm 1, when event k is being processed, data for events τ>k have not yet been observed, so the integral over 0..K is not computable without using future-event statistics. The paper does not state whether the integral is evaluated over all τ (which would leak future information into the dynamics model) or only over τ≤k (which would make Eq. (12) as written not the actual objective). This ambiguity is load-bearing: the ablation in Table 4 shows that removing the environmental feature causes the largest drop (e.g., -2.53 on Twitter), so a substantial part of the reported gain is attributed to this module. Please specify the exact computation, and if future statistics are used, rerun the experiments with a causally valid objective; if only past/current statistics are used, rewrite Eq. (12) and the surrounding text accordingly.
  2. [§3.2, Eqs. (4)-(6)] The Dirichlet process mixture is allegedly implemented by the responsibility formula, but the notation Σρ_{<i,m} is undefined: it is not stated whether this is a sum over previous samples' responsibilities for expert m, how it is accumulated across batches, or how it is updated when a new expert is created. In addition, the 'variational generation loss' L_VG is implemented as the reconstruction loss ‖z_i E_m^G − z_i‖₂², which is not a variational lower bound, and the responsibilities are not normalized. The text calls this a Dirichlet process-based approach, but the connection to a Dirichlet process is not established. Please either provide a formal derivation with the exact update rules, or revise the terminology to describe the heuristic it actually is. This matters because the 'remember past' contribution is supported by this module.
  3. [§4.2, Table 2] The caption states that the bold results are 'statistically significant than the baseline models,' but no significance test is described anywhere in the paper or supplementary. Several comparisons have overlapping standard-deviation intervals (e.g., GossipCop F1_fake for CAFE+DaedCmd is 61.05±0.8 vs 59.21±1.5, with intervals that overlap), so the claim of significance as used in the abstract needs a formal test. Please report the test procedure (e.g., paired bootstrap or Wilcoxon signed-rank test over the five seeds) and the resulting p-values or confidence intervals.
  4. [§4.2, Table 3 and text] The text states that 'the Twitter dataset, being the smallest in scale, achieves the largest improvement,' but Table 3 shows Twitter has 13,924 articles, which is larger than GossipCop (12,840) and Weibo (9,528). If 'scale' refers to the number of unique images (Twitter has 514), that should be stated explicitly; as written, the sentence is internally inconsistent and the explanation of the ordering of improvements needs to be corrected.
minor comments (6)
  1. [Algorithm 1, step 13] Step 13 says 'Construct the distribution P(ê|y=1) with Eq.(10)', but Eq. (10) defines the ODE derivatives; the distribution parameters are computed in Eq. (9). Please correct the reference.
  2. [§3.3, Eq. (12)] The integral in Eq. (12) is written with a 'dt' but the integration variable should be τ (the temporal index). The notation is inconsistent and should be unified.
  3. [§3.2, Eq. (7)] The routing weight r_i is written as z_i W_R, but the dimension of W_R is not specified; please clarify that it outputs a scalar, or describe the output dimension if it is a vector.
  4. [General] The method name is rendered inconsistently as 'DaedCmd' and 'DAEDCMD' across the abstract and body; please unify the spelling.
  5. [Table 4] The column 'Avg.↓' is not defined; please state how the average decline is computed and whether it is averaged over the five metrics listed.
  6. [Section 1, footnote] The GitHub URL in the introduction contains spaces ('https:// github.com/ wangbing1416/ DAEDCMD'); please format it as a proper link.

Circularity Check

2 steps flagged · score 6.0 of 10

Eq. (12) trains the dynamics model on future event distributions, making the 'anticipate future' gains a leakage artifact rather than held-out prediction.

  1. fitted input called prediction [Section 3.3, Eq. (12); Appendix A, Algorithm 1 (loop over k=1..K, L_DM computed at line 14)]
    "Following Eq. (9), we compute the mean μ_τ and covariance σ_τ of each batched data whose t_i = τ as the ground-truth label, and formulate the following objective: L_DM = 1/K ∫_0^K (||μ̂_τ − μ_τ||^2 + ||σ̂_τ − σ_τ||^2) dt."

    In Algorithm 1, training proceeds sequentially over events k = 1..K, yet Eq. (12) defines L_DM as an integral over all τ ∈ [0,K] against 'ground-truth' μ_τ and σ_τ computed for every time period. At the moment event k is being trained, the distributions for τ > k are future data. Unless the integral is silently restricted to τ ≤ k, the dynamics model is fit to the very future distributions it is then claimed to 'predict,' so the ablation gains attributed to the 'anticipate future' module (Table 4, w/o Feature ê) are leakage rather than held-out prediction. If the restriction is intended, Eq. (12) is not the actual objective, and the paper does not say so.

  2. self definitional [Section 3.3, Eq. (9); Eq. (2), Eq. (3)]
    "we first use the multimodal feature e^{y=1} of fake articles to build a environmental distribution specified by a Gaussian distribution P(ê|y=1)=N(μ, σ²)... Given the multimodal feature e_i from the dynamically adapted MoE module, we map the samples with their labels y=1 to a shared environmental space and compute their mean and covariance matrix... Based on the multimodal feature e_i and the sampled environmental feature ê_i, the veracity prediction objective is as follows: L_VP=..."

    The 'environmental distribution' used as ground truth is not an independently observed social distribution; Eq. (9) defines it as the mean and covariance of the model's own features e_i. The dynamics model is then trained to reproduce these internal statistics, and the sample ê_i drawn from the predicted distribution is concatenated to e_i for the same classifier. Thus the predicted environmental feature is fed back into the very representation whose statistics define the target, so the module is fitted to a model-generated quantity rather than an external environment.

full rationale

Most of the empirical comparison is not circular: DaedCmd is evaluated on held-out future articles against six MMD baselines and three continual learning baselines, and the MoE/DP expert expansion is an independent parameter-isolation mechanism. The paper's self-citations (e.g., [32-34]) are related-work references and are not load-bearing. However, the paper's second stated contribution, anticipating future environmental distributions, is compromised by Eq. (12) as written: the dynamics-model loss integrates over all time periods up to K, using 'ground-truth' mean and covariance computed from future batches, while Algorithm 1 sequentially processes events. This makes the 'prediction' a fit to the future distributions the model is supposed to forecast, so the ablation gains for Feature ê do not demonstrate held-out anticipation. Additionally, the environmental distribution is defined from the model's own features, closing an internal loop. Because the end-task benchmark is still externally evaluated, the circularity is partial, not total.

Assumptions & free parameters 6 free parameters · 4 assumptions · 1 invented entities

The method rests on several hand-chosen hyperparameters (alpha, beta, gamma, lambda, epsilon, xi) and on domain assumptions about the Gaussian shape of the fake-article distribution and the use of event index as continuous time. The Dirichlet process component is best understood as a heuristic expansion rule rather than a formal probabilistic model. The only invented entity is the environmental feature, which has no independent evidence.

free parameters (6)
  • alpha = 0.1
    Trade-off weight for the contrastive loss L_CL in Eq. (3); fixed by hand.
  • beta = 0.1
    Trade-off weight for the variational/generation loss L_VG in Eq. (3); fixed by hand.
  • gamma = 1
    Trade-off weight for the dynamics model loss L_DM; sensitivity analysis (Fig. 5) shows gamma=1 is optimal on all datasets, chosen by hand.
  • lambda = not specified
    Sensitivity to new data in the Dirichlet process responsibility (Eq. (4)); no value is reported.
  • epsilon = 0.99
    EMA smoothing parameter for the event-shared expert (Eq. (8)); empirically fixed.
  • xi = not specified
    Temperature in the contrastive learning objective (Eq. (1)); no value is reported.
assumptions (4)
  • domain assumption The environmental distribution of fake articles is Gaussian, P(e_hat|y=1)=N(mu,sigma^2), in the learned feature space (Eq. (9)).
    This Gaussian assumption is central to the dynamics model; no justification is given for why fake-post features should be Gaussian.
  • ad hoc to paper Event index k can be used as continuous time t in the ODE dynamics (Eq. (12) integrates over 0..K).
    The mapping of discrete event subsets to a continuous time axis with uniform spacing is a modeling choice, not derived from the data.
  • ad hoc to paper The Dirichlet process expansion is implemented by the responsibility formula in Eq. (4) with a pseudo-count term -log sum rho and a reconstruction loss L_VG; this is treated as a Dirichlet process mixture in the text.
    The formula is not the posterior of a Dirichlet process mixture model; it is an ad hoc scoring heuristic (the generator part is described as a VAE but L_VG is only a reconstruction objective).
  • standard math Neural ODE solvers (dopri5) provide a faithful solution to Eq. (10).
    Standard numerical ODE integration is invoked for Eq. (13).
invented entities (1)
  • Environmental feature e_hat
    purpose: A latent feature sampled from the predicted Gaussian environmental distribution and concatenated to the MoE feature for veracity classification (Eq. (2)).
    It is an internal latent variable defined entirely by the model's architecture and training; there is no external measurement or falsifiable handle to validate the 'environment' it is meant to represent.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Remember Past, Anticipate Future: Learning Continual Multimodal Misinformation Detectors." pith.science (2026). https://pith.science/paper/3JIK4NAZ

@misc{pith2026250705939,
  author       = {Pith},
  title        = {Pith review of: Remember Past, Anticipate Future: Learning Continual Multimodal Misinformation Detectors},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3JIK4NAZ}},
  note         = {Machine review of arXiv:2507.05939}
}
read the original abstract

Nowadays, misinformation articles, especially multimodal ones, are widely spread on social media platforms and cause serious negative effects. To control their propagation, Multimodal Misinformation Detection (MMD) becomes an active topic in the community to automatically identify misinformation. Previous MMD methods focus on supervising detectors by collecting offline data. However, in real-world scenarios, new events always continually emerge, making MMD models trained on offline data consistently outdated and ineffective. To address this issue, training MMD models under online data streams is an alternative, inducing an emerging task named continual MMD. Unfortunately, it is hindered by two major challenges. First, training on new data consistently decreases the detection performance on past data, named past knowledge forgetting. Second, the social environment constantly evolves over time, affecting the generalization on future data. To alleviate these challenges, we propose to remember past knowledge by isolating interference between event-specific parameters with a Dirichlet process-based mixture-of-expert structure, and anticipate future environmental distributions by learning a continuous-time dynamics model. Accordingly, we induce a new continual MMD method DAEDCMD. Extensive experiments demonstrate that DAEDCMD can consistently and significantly outperform the compared methods, including six MMD baselines and three continual learning methods.

Figures

Figures reproduced from arXiv: 2507.05939 by the authors.

Figure 1
Figure 1. Preliminary experiments that demonstrate Macro [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overall Framework of DaedCmd. Given the text x 𝑡 and image x 𝑣 , we first extract their semantics and fuse them as a multimodal feature z. To mitigate forgetting, we propose a dynamically adapted MoE to obtain the forgetting resistant feature e. To learn the dynamically evolved environmental distribution, we construct it as a Gaussian distribution, and learn a dynamics model to predict the environmental distribution… view at source ↗
Figure 3
Figure 3. The changes of the model’s accuracy on other event subsets when it is incrementally trained on event subsets. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: The changes of the model’s accuracy on the first [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 3
Figure 3. Figure 3: Additionally, to provide a more concrete visualization of the [PITH_FULL_IMAGE:figures/full_fig_p008_3.png]
Figure 5
Figure 5. Figure 5: Sensitivity analysis of the parameter 𝛾. In Michigan: Woman enters Guinness book of record after 14 ch￾ildren from 14 differ￾ent fathers. [ Feb 24 20:07:09 2016 ] Sharks seen roami - ng in New Jersey streets and metro stations. #Sandy [ Nov 01 11:27:27 2012 ] Future Pa…
Figure 6
Figure 6. Figure 6: Case study that demonstrates the fake class pre [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

52 extracted references · 49 canonical work pages

  1. [1]

    Christina Boididou, Symeon Papadopoulos, Markos Zampoglou, Lazaros Aposto- lidis, Olga Papadopoulou, and Yiannis Kompatsiaris. 2018. Detection and Visual- ization of Misleading Content on Twitter. International Journal of Multimedia Information Retrieval 7, 1 (2018), 71–86

    work page 2018

  2. [2]

    Tian Qi Chen, Yulia Rubanova, Jesse Bettencourt, and David Duvenaud. 2018. Neural Ordinary Differential Equations. In Advances in Neural Information Pro- cessing Systems. 6572–6583

    work page 2018

  3. [3]

    Yixuan Chen, Dongsheng Li, Peng Zhang, Jie Sui, Qin Lv, Tun Lu, and Li Shang

    work page

  4. [4]

    Cyprien de Masson d’Autume, Sebastian Ruder, Lingpeng Kong, and Dani Yo- gatama. 2019. Episodic Memory in Lifelong Language Learning. In Advances in Neural Information Processing Systems . 13122–13131

    work page 2019

  5. [5]

    Michela Del Vicario, Alessandro Bessi, Fabiana Zollo, Fabio Petroni, Antonio Scala, Guido Caldarelli, H Eugene Stanley, and Walter Quattrociocchi. 2016. The Spreading of Misinformation Online. Proceedings of the national academy of Sciences 113, 3 (2016), 554–559

    work page 2016

  6. [6]

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. In Conference of the North American Chapter of the Association for Computational Linguistics. 4171–4186

    work page 2019

  7. [7]

    Yasan Ding, Bin Guo, Yan Liu, Yao Jing, Maolong Yin, Nuo Li, Hao Wang, and Zhiwen Yu. 2025. EvolveDetector: Towards an Evolving Fake News Detector for Emerging Events with Continual Knowledge Accumulation and Transfer. Information Processing & Management 62, 1 (2025), 103878

    work page 2025

  8. [8]

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xi- aohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, and Neil Houlsby. 2021. An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale. InInternational Conference on Learning Representations

    2021

Show all 52 references
  1. [9]

    Shihan Dou, Enyu Zhou, Yan Liu, Songyang Gao, Wei Shen, Limao Xiong, Yuhao Zhou, Xiao Wang, Zhiheng Xi, Xiaoran Fan, Shiliang Pu, Jiang Zhu, Rui Zheng, Tao Gui, Qi Zhang, and Xuanjing Huang. 2024. LoRAMoE: Alleviating World Knowledge Forgetting in Large Language Models via MoE...

    2024

  2. [10]

    Wenfeng Feng, Chuzhan Hao, Yuewei Zhang, Yu Han, and Hao Wang. 2024. Mixture-of-LoRAs: An Efficient Multitask Tuning Method for Large Language Models. In International Conference on Computational Linguistics . 11371–11380

    2024

  3. [11]

    Robert M. French. 1993. Catastrophic Interference in Connectionist Networks: Can It Be Predicted, Can It Be Prevented?. In Advances in Neural Information Processing Systems. 1176–1177

    1993

  4. [12]

    Yi Han, Shanika Karunasekera, and Christopher Leckie. 2021. Continual Learn- ing for Fake News Detection from Social Media. In International Conference on Artificial Neural Networks, Vol. 12892. 372–384

    2021

  5. [13]

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep Residual Learning for Image Recognition. In IEEE Conference on Computer Vision and Pattern Recognition. 770–778

    2016

  6. [14]

    Beizhe Hu, Qiang Sheng, Juan Cao, Yongchun Zhu, Danding Wang, Zhengjia Wang, and Zhiwei Jin. 2023. Learn over Past, Evolve for Future: Forecasting Temporal Trends for Fake News Detection. In Annual Meeting of the Association for Computational Linguistics: Industry Track . 116–125

    2023

  7. [15]

    Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen

    Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2022. LoRA: Low-Rank Adaptation of Large Language Models. In International Conference on Learning Representations

    2022

  8. [16]

    Jianheng Huang, Leyang Cui, Ante Wang, Chengyi Yang, Xinting Liao, Linfeng Song, Junfeng Yao, and Jinsong Su. 2024. Mitigating Catastrophic Forgetting in Large Language Models with Self-Synthesized Rehearsal. In Annual Meeting of the Association for Computational Linguistics ....

    2024

  9. [17]

    Belongie

    Xun Huang and Serge J. Belongie. 2017. Arbitrary Style Transfer in Real-Time with Adaptive Instance Normalization. In IEEE International Conference on Com- puter Vision. 1510–1519

    2017

  10. [18]

    Zhiwei Jin, Juan Cao, Han Guo, Yongdong Zhang, and Jiebo Luo. 2017. Multimodal Fusion with Recurrent Neural Networks for Rumor Detection on Microblogs. In ACM on Multimedia Conference. 795–816

    2017

  11. [19]

    Zhiwei Jin, Juan Cao, Yu-Gang Jiang, and Yongdong Zhang. 2014. News Credi- bility Evaluation on Microblog with a Hierarchical Propagation Model. In IEEE International Conference on Data Mining . 230–239

    2014

  12. [20]

    Kingma and Max Welling

    Diederik P. Kingma and Max Welling. 2014. Auto-Encoding Variational Bayes. In International Conference on Learning Representations

    2014

  13. [21]

    Rabinowitz, Joel Veness, Guillaume Desjardins, Andrei A

    James Kirkpatrick, Razvan Pascanu, Neil C. Rabinowitz, Joel Veness, Guillaume Desjardins, Andrei A. Rusu, Kieran Milan, John Quan, Tiago Ramalho, Agnieszka Grabska-Barwinska, Demis Hassabis, Claudia Clopath, Dharshan Kumaran, and Raia Hadsell. 2016. Overcoming Catastrophic For...

    2016 arXiv

  14. [22]

    Nayeon Lee, Andrea Madotto, Yejin Bang, and Pascale Fung. 2021. Dynamically Addressing Unseen Rumor via Continual Learning. CoRR abs/2104.08775 (2021)

    2021 arXiv

  15. [23]

    Soochan Lee, Junsoo Ha, Dongsu Zhang, and Gunhee Kim. 2020. A Neural Dirich- let Process Mixture Model for Task-Free Continual Learning. In International Conference on Learning Representations

    2020

  16. [24]

    David Lopez-Paz and Marc’Aurelio Ranzato. 2017. Gradient Episodic Memory for Continual Learning. In Advances in Neural Information Processing Systems . 6467–6476

    2017

  17. [25]

    Yufei Ma, Zihan Liang, Huangyu Dai, Ben Chen, Dehong Gao, Zhuoran Ran, Zihan Wang, Linbo Jin, Wen Jiang, Guannan Zhang, Xiaoyan Cai, and Libin Yang

  18. [26]

    Mark B. Ring. 1997. CHILD: A First Step Towards Continual Learning. Machine Learning 28, 1 (1997), 77–104

    1997

  19. [27]

    Anthony V. Robins. 1995. Catastrophic Forgetting, Rehearsal and Pseudorehearsal. Connection Science 7, 2 (1995), 123–146

    1995

  20. [28]

    Gobinda Saha, Isha Garg, and Kaushik Roy. 2021. Gradient Projection Memory for Continual Learning. In International Conference on Learning Representations

    2021

  21. [29]

    Zhejie Shao, Guoyong Cai, Qinghua Liu, and Yunxian Shang. 2024. An Active Learning Framework for Continuous Rapid Rumor Detection in Evolving Social Media. In International Joint Conference on Neural Networks . 1–8

    2024

  22. [30]

    Lingzhi Shen, Yunfei Long, Xiaohao Cai, Imran Razzak, Guanming Chen, Kang Liu, and Shoaib Jameel. 2025. GAMED: Knowledge Adaptive Multi-Experts De- coupling for Multimodal Fake News Detection. In ACM International Conference on Web Search and Data Mining . 586–595

    2025

  23. [31]

    Kai Shu, Deepak Mahudeswaran, Suhang Wang, Dongwon Lee, and Huan Liu

  24. [32]

    Bing Wang, Ximing Li, Changchun Li, Bo Fu, Songwen Pei, and Shengsheng Wang. 2024. Why Misinformation is Created? Detecting them by Integrating Intent Features. In ACM International Conference on Information and Knowledge Management. 2304–2314

    2024

  25. [33]

    Bing Wang, Ximing Li, Changchun Li, Shengsheng Wang, and Wanfu Gao. 2024. Escaping the Neutralization Effect of Modality Features Fusion in Multimodal Fake News Detection. Information Fusion 111 (2024), 102500

    2024

  26. [34]

    Bing Wang, Shengsheng Wang, Changchun Li, Renchu Guan, and Ximing Li

  27. [35]

    Longzheng Wang, Chuang Zhang, Hongbo Xu, Yongxiu Xu, Xiaohan Xu, and Siqi Wang. 2023. Cross-modal Contrastive Learning for Multimodal Fake News Detection. In ACM International Conference on Multimedia . 5696–5704

    2023

  28. [36]

    Xiao Wang, Tianze Chen, Qiming Ge, Han Xia, Rong Bao, Rui Zheng, Qi Zhang, Tao Gui, and Xuanjing Huang. 2023. Orthogonal Subspace Learning for Lan- guage Model Continual Learning. InFindings of the Association for Computational Linguistics: EMNLP. 10658–10671

    2023

  29. [37]

    Yaqing Wang, Fenglong Ma, Zhiwei Jin, Ye Yuan, Guangxu Xun, Kishlay Jha, Lu Su, and Jing Gao. 2018. EANN: Event Adversarial Neural Networks for Multi-Modal Fake News Detection. In ACM SIGKDD International Conference on Knowledge Discovery & Data Mining . 849–857

    2018

  30. [38]

    In ACM International Conference on Multimedia

    Harmfully Manipulated Images Matter in Multimodal Misinformation Detection. In ACM International Conference on Multimedia . 2262–2271

  31. [39]

    Gerhard Wanner and Ernst Hairer. 1996. Solving ordinary differential equations II . Vol. 375

    1996

  32. [40]

    Yang Wu, Pengwei Zhan, Yunjian Zhang, LiMing Wang, and Zhen Xu. 2021. Multimodal Fusion with Co-Attention Networks for Fake News Detection. In Findings of the Association for Computational Linguistics: ACL/IJCNLP. 2560–2569

    2021

  33. [41]

    Fei Ye and Adrian G. Bors. 2021. Lifelong Infinite Mixture Model Based on Knowledge-Driven Dirichlet Process. In IEEE/CVF International Conference on Computer Vision. 10675–10684

    2021

  34. [42]

    Yaqing Wang, Fenglong Ma, Haoyu Wang, Kishlay Jha, and Jing Gao. 2021. Multimodal Emergent Fake News Detection via Meta Neural Process Networks. In ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 3708–3716

    2021

  35. [43]

    Chengxi Zang and Fei Wang. 2020. Neural Dynamics on Complex Networks. In ACM SIGKDD Conference on Knowledge Discovery and Data Mining . 892–902

    2020

  36. [44]

    Jiajun Zhang, Zhixun Li, Qiang Liu, Shu Wu, Zilei Wang, and Liang Wang

  37. [45]

    Litian Zhang, Xiaoming Zhang, Ziyi Zhou, Feiran Huang, and Chaozhuo Li. 2024. Reinforced Adaptive Knowledge Learning for Multimodal Fake News Detection. In AAAI Conference on Artificial Intelligence . 16777–16785

    2024

  38. [46]

    Qichao Ying, Xiaoxiao Hu, Yangming Zhou, Zhenxing Qian, Dan Zeng, and Shiming Ge. 2023. Bootstrapping Multi-View Representations for Fake News Detection. In AAAI Conference on Artificial Intelligence . 5384–5392

    2023

  39. [47]

    Yuhui Zuo, Wei Zhu, and Guoyong Cai. 2022. Continually Detection, Rapidly React: Unseen Rumors Detection Based on Continual Prompt-Tuning. In Interna- tional Conference on Computational Linguistics . 3029–3041. MM ’25, October 27–31, 2025, Dublin, Ireland Bing Wang, Ximing Li,...

    2022

  40. [49]

    In ACM International Conference on Information and Knowledge Management

    Evolving to the Future: Unseen Event Adaptive Fake News Detection on Social Media. In ACM International Conference on Information and Knowledge Management. 4273–4277

  41. [51]

    Xinyi Zhou, Jindi Wu, and Reza Zafarani. 2020. SAFE: Similarity-Aware Multi- modal Fake News Detection. InAdvances in Knowledge Discovery and Data Mining - Pacific-Asia Conference (Lecture Notes in Computer Science, Vol. 12085) . 354–367

    2020

  42. [2020]

    Big Data 8, 3 (2020), 171–188

    FakeNewsNet: A Data Repository with News Content, Social Context, and Spatiotemporal Information for Studying Fake News on Social Media. Big Data 8, 3 (2020), 171–188

    2020

  43. [2022]

    In The ACM Web Conference

    Cross-modal Ambiguity Learning for Multimodal Fake News Detection. In The ACM Web Conference. 2897–2905

  44. [2024]

    In Conference on Empirical Methods in Natural Language Processing

    MoDULA: Mixture of Domain-Specific and Universal LoRA for Multi-Task Learning. In Conference on Empirical Methods in Natural Language Processing . 2758–2770

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.