Pith. sign in

REVIEW 3 major objections 4 minor 101 references

This paper reports that a multimodal BiLSTM-AM-VMD architecture achieves 0.963 AUC and 0.910 F1 for early liver cancer diagnosis on a 648-patient dataset, with an ablation showing attention and VMD each contribute.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-05 12:48 UTC pith:DA76O4OQ

load-bearing objection The architecture is a plausible combination of known pieces, but the dataset descriptions are for entirely different diseases, so the reported liver-cancer results are unsupported. the 3 major comments →

arxiv 2509.01164 v1 pith:DA76O4OQ submitted 2025-09-01 cs.LG eess.IV

A Multimodal Deep Learning Framework for Early Diagnosis of Liver Cancer via Optimized BiLSTM-AM-VMD Architecture

Cheng Cheng , Zeping Chen , Xavier Wang This is my paper
classification cs.LG eess.IV
keywords hepatocellular carcinomaearly diagnosismultimodal deep learningBiLSTMmulti-head attentionvariational mode decompositionparticle swarm optimizationsurvival analysis
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper aims to establish that one architecture—bidirectional LSTM sequence modeling, multi-head attention, and variational mode decomposition, tuned by particle swarm optimization—can diagnose early liver cancer from heterogeneous patient data better than standard machine-learning baselines. On a dataset the paper calls real-world, 648 patients with demographic, hormonal, metabolic, and imaging-derived features, it reports an AUC of 0.963, F1 of 0.910, sensitivity of 0.921, and specificity of 0.907, beating random forest, XGBoost, and a plain LSTM. The ablation study says attention and VMD each add measurable performance, and a second survival analysis reports C-index 0.821 and one-year AUC 0.876 for recurrence-free survival. The intended payoff is clinical: early hepatocellular carcinoma is often silent and alpha-fetoprotein is an imperfect marker, so a multimodal model that reliably flags early-stage patients could shift diagnosis earlier. The paper also offers attention weights and SHAP values as evidence that the model points to the same risk factors clinicians already use.

Core claim

The central claim is that combining multiscale signal decomposition with bidirectional temporal encoding and attention-based feature weighting is enough to lift early-HCC prediction above standard baselines on the same data. In the proposed pipeline, variational mode decomposition splits each noisy input feature into frequency sub-signals; a bidirectional LSTM reads those sub-signals in both directions to capture forward and backward dependencies; multi-head self-attention then weights the most informative time steps and features; and a particle-swarm optimizer selects hyperparameters by maximizing validation AUC. The paper reports that this combination reaches 0.963 AUC and 0.910 F1 on the

What carries the argument

The load-bearing object is the BiLSTM-AM-VMD pipeline itself: VMD (variational mode decomposition) splits a variable into intrinsic mode functions, i.e., sub-signals representing distinct frequency bands; a BiLSTM, a recurrent network reading sequences in both directions, encodes temporal or pseudo-temporal structure; multi-head attention weights features dynamically, aiding interpretability; and PSO (particle swarm optimization) searches hyperparameters. Each part is meant to handle a known failure: VMD damps noise in biomarkers, BiLSTM captures dependence across modalities and visits, attention highlights salient features, and PSO removes manual tuning. The paper's proof strategy is compar

Load-bearing premise

The load-bearing premise is that the 648-patient dataset is genuinely a liver-cancer cohort with the features and outcomes described in Sections V-A and VII-A; if the cohort is actually a different disease population, every reported accuracy and ranking loses its meaning.

What would settle it

Request the dataset specification and cohort labels. If the features are from a polycystic ovary syndrome workup or the survival cohort carries glioma markers (IDH, MGMT, Ki-67), the liver-cancer claim fails. Alternatively, run the trained model on an externally confirmed HCC cohort: if its AUC falls to the level of the baselines, the reported 0.963 is not a general early-liver-cancer measurement.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • If the reported numbers replicate, a multimodal pipeline of this shape can rank early-HCC risk from routine demographics, laboratory values, and imaging-derived features without hand-crafted feature engineering.
  • The ablation pattern—attention mainly raises sensitivity, VMD mainly adds robustness to noisy signals—gives a concrete recipe for other cancer-screening tasks with heterogeneous data.
  • Extending the same architecture to survival prediction means one model family could cover both diagnosis and recurrence-risk stratification.
  • The attention and SHAP outputs identify hormone ratios and imaging texture features as the most influential inputs, suggesting candidate biomarkers for prospective studies.
  • PSO-based hyperparameter tuning makes the pipeline adaptable to new datasets at the cost of extra compute.

Where Pith is reading between the lines

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

  • The reported accuracies are only as meaningful as the dataset label. The feature list in Section V-A (menstrual irregularity, hirsutism, LH/FSH, follicle counts) and the survival cohort markers in Section VII-A (IDH mutation, MGMT methylation, Ki-67) are markers of other disease profiles, so the single most informative check is whether the cohort is actually liver cancer.
  • The pseudo-temporal sequences built by grouping feature modalities are a weak stand-in for true longitudinal records; a test on real repeated measurements would better judge whether BiLSTM's bidirectional temporal modeling is doing the work or whether cross-sectional feature ordering inflates its role.
  • The manuscript contains a self-description as a revised version with citations inserted from a provided list; a careful reader should treat the reference list as a provenance signal rather than as evidence of a coherent research lineage.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes BiLSTM-AM-VMD, a multimodal deep learning framework for early diagnosis of hepatocellular carcinoma (HCC). The model combines a bidirectional LSTM, multi-head attention, and variational mode decomposition, with PSO-based hyperparameter tuning. The authors report strong empirical results on a claimed real-world clinical dataset of 648 samples: AUC 0.963 and F1 0.910 for early HCC classification (Section VI, Table I), and a C-index of 0.821 with 1-year AUC 0.876 for recurrence-free survival prediction (Section VII, Table IV). The manuscript contains two separate 'Results' sections (VI and VII) that describe incompatible tasks and contradictory cohort characteristics. Section V-A lists features typical of polycystic ovary syndrome (menstrual irregularity, hirsutism, LH/FSH, follicle counts), while Section VII-A describes a cohort with IDH mutation, MGMT methylation, and Ki-67, which are markers characteristic of glioma. These internal inconsistencies make the reported performance numbers uninterpretable and the central claim unsupported.

Significance. If the proposed architecture were evaluated on a coherent, real-world HCC cohort and outperformed strong baselines with proper uncertainty quantification, the result would be a useful contribution to multimodal medical machine learning. The combination of VMD, BiLSTM, and attention is a plausible engineering contribution, and the paper explicitly describes PSO-based hyperparameter search and ablation studies. However, the empirical claim is load-bearing and collapses under the manuscript's own dataset descriptions. The paper provides no code, no data, and no external validation, and it contains an inserted meta-comment indicating automated text assembly. The significance of the claimed results cannot be assessed because the evaluation protocol is internally contradictory; in its current form the paper does not establish that any liver cancer prediction was performed.

major comments (3)
  1. [§V-A, §VI-A, Table I] The dataset described in Section V-A is not a liver cancer cohort. The listed features — menstrual irregularity, hirsutism score, acne, alopecia, LH, FSH, testosterone, SHBG, LH/FSH ratio, fasting glucose/insulin, HOMA-IR, and follicle counts — are characteristic of polycystic ovary syndrome, not hepatocellular carcinoma. No HCC-specific variables such as AFP, liver enzymes, cirrhosis status, or liver imaging features are mentioned. Despite this, Table I reports an AUC of 0.963 for 'early HCC diagnosis.' If the data are not from a liver cancer cohort, the classification results are vacuous.
  2. [§VII-A, §VII-C, Tables III and IV] Section VII introduces a completely different set of clinical variables for the same N=648: median age 56.4, tumor size, IDH mutation (14.7%), MGMT methylation (46.5%), Ki-67, MRI texture entropy, and recurrence-free survival endpoints. IDH mutation and MGMT methylation are canonical glioma markers, not liver cancer markers. Furthermore, Section VI evaluates binary classification (AUC, F1, sensitivity, specificity), while Section VII evaluates time-to-event prediction (C-index, time-dependent AUC). The two 'Results' sections describe incompatible tasks and incompatible feature sets; the reported C-index of 0.821 and 1-year AUC of 0.876 have no demonstrable connection to the classification experiments or to liver cancer.
  3. [§I-A, §I-C] The paper contains an explicit meta-comment in Section I-A: 'The revised version maintains the original scientific content while incorporating relevant citations from the provided list in a logical manner.' This is not scientific prose; it is a trace of an automated text-generation/revision pipeline. The organization statement in Section I-C is also inaccurate (it states Section V presents results and Section VII reports results, which does not match the actual structure). These issues are not merely typographical: they compound the dataset contradictions and prevent the reader from trusting that the manuscript was assembled with care. A scientific paper must present a single coherent study; this manuscript does not.
minor comments (4)
  1. [§III-D, §V-B] Variational mode decomposition is applied to 'pseudo-temporal sequences' constructed by grouping features by modality in a single-visit dataset. VMD is a signal-processing method that assumes a meaningful frequency-domain representation. Applying it to arbitrarily ordered tabular features is not justified, and the paper does not explain what 'frequency' means for clinical variables such as hormone levels or imaging-derived metrics.
  2. [Tables I, II, IV] No confidence intervals, standard deviations, or statistical significance tests are reported for any metric. With N=648 and 5-fold cross-validation, the differences between models (e.g., AUC 0.963 vs 0.922) need uncertainty quantification. The paper also lacks external validation on an independent cohort, which is essential for a clinical prediction claim.
  3. [Algorithm 1, §III-E] The PSO hyperparameter optimization is described only at a high level. Population size, number of iterations, search ranges for hyperparameters, and the final selected hyperparameter values are not reported, so the experiments are not reproducible.
  4. [References] The reference list contains a large block of citations in the introduction ([1]–[68]) that are unrelated to liver cancer or the proposed method. The citation list appears padded rather than focused on the clinical and methodological context.

Circularity Check

0 steps flagged

No significant circularity: the predicted metrics are empirical outputs of a standard supervised pipeline, not consequences of the model definition; the dataset/label inconsistency is a data-integrity concern, not a circular-reasoning reduction.

full rationale

The claimed derivation chain is not circular. BiLSTM, multi-head attention, VMD, and PSO are defined independently of the label; the model is trained with binary cross-entropy and evaluated with 5-fold CV, so the reported AUC/F1 are empirical fits rather than quantities that equal the inputs by construction. Hyperparameters chosen by PSO on validation AUC and then evaluated on test folds is standard practice and does not rename a fitted parameter as a prediction. The extensive references and self-citations in the introduction are contextual padding, not load-bearing justification for the architecture or the empirical results. The paper's own limitations (Section VIII) acknowledge missing-data and temporal-alignment issues; these affect validity, not circularity. The most serious weakness is the apparent mismatch between the disease label (liver cancer) and the feature set (Section V-A lists PCOS markers such as menstrual irregularity, hirsutism, LH/FSH, and follicle counts; Section VII-A lists IDH/MGMT/Ki-67 and survival endpoints typical of glioma). This is a data-validity/reproducibility problem that would make the results uninterpretable, but it is not an instance of the derivation reducing to its own inputs by construction. The inserted AI-generated meta-comment in Section I-A is a writing artifact and does not create a circular step. Therefore, under the defined circularity criteria, the appropriate score is 0.

Axiom & Free-Parameter Ledger

4 free parameters · 3 axioms · 0 invented entities

The paper's central claim rests on standard deep learning components and an unreleased dataset. The only quantities tuned by the authors are hyperparameters selected via PSO to maximize validation AUC; these are fitted values. The dataset itself is an unexplained, unreleased collection with contradictory feature descriptions, and the pseudo-temporal ordering is an ad hoc modeling choice.

free parameters (4)
  • LSTM hidden size
    Chosen by PSO to maximize validation AUC. Not reported in the paper.
  • Number of attention heads
    Chosen by PSO to maximize validation AUC. Not reported in the paper.
  • Number of VMD modes K
    Chosen by PSO to maximize validation AUC. Not reported in the paper.
  • Dropout rate
    Chosen by PSO to maximize validation AUC. Not reported in the paper.
axioms (3)
  • domain assumption The dataset of 648 patients is a genuine liver cancer cohort with the described multimodal features.
    The entire empirical evaluation depends on this, but the feature lists in Section V-A and VII-A are mutually inconsistent and characteristic of other diseases.
  • domain assumption Grouping features by modality creates meaningful pseudo-temporal sequences for BiLSTM.
    The paper acknowledges data were collected in single-visit format and constructs pseudo-temporal sequences by feature type. This is an ad hoc modeling choice with no justification that it captures temporal structure.
  • domain assumption VMD decomposition of the tabular/pseudo-temporal features yields useful frequency components for classification.
    VMD is normally applied to time series signals; applying it to concatenated tabular features without validation is an unsupported assumption.

pith-pipeline@v1.4.0-alltime-deepseek-medium · 13921 in / 11084 out tokens · 112701 ms · 2026-08-05T12:48:28.747732+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of A Multimodal Deep Learning Framework for Early Diagnosis of Liver Cancer via Optimized BiLSTM-AM-VMD Architecture." pith.science (2026). https://pith.science/paper/DA76O4OQ

@misc{pith2026250901164,
  author       = {Pith},
  title        = {Pith review of: A Multimodal Deep Learning Framework for Early Diagnosis of Liver Cancer via Optimized BiLSTM-AM-VMD Architecture},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DA76O4OQ}},
  note         = {Machine review of arXiv:2509.01164}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

This paper proposes a novel multimodal deep learning framework integrating bidirectional LSTM, multi-head attention mechanism, and variational mode decomposition (BiLSTM-AM-VMD) for early liver cancer diagnosis. Using heterogeneous data that include clinical characteristics, biochemical markers, and imaging-derived variables, our approach improves both prediction accuracy and interpretability. Experimental results on real-world datasets demonstrate superior performance over traditional machine learning and baseline deep learning models.

Figures

Figures reproduced from arXiv: 2509.01164 by Cheng cheng, Xavier Wang, Zeping Chen.

Figure 2
Figure 2. Figure 2: Confusion matrix of the BiLSTM-AM-VMD model. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 1
Figure 1. Figure 1: ROC curves for selected models. The proposed model shows superior [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 4
Figure 4. Figure 4: Calibration curve and decision curve analysis (DCA) of BiLSTM [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 3
Figure 3. Figure 3: Univariate Cox regression forest plot for top features [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 6
Figure 6. Figure 6: Kaplan–Meier curves for model-based risk groups [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

101 extracted references · 37 canonical work pages · 1 internal anchor

  1. [1]

    mplug-owl: Modularization empowers large language models with multimodality,

    Q. Ye, H. Xu, G. Xu, J. Ye, M. Yan, Y . Zhou, J. Wang, A. Hu, P. Shi, Y . Shi et al. , “mplug-owl: Modularization empowers large language models with multimodality,” arXiv preprint arXiv:2304.14178 , 2023

    Pith/arXiv arXiv 2023

  2. [2]

    Analyzing and mitigating object hallucination in large vision-language models,

    Y . Zhou, C. Cui, J. Yoon, L. Zhang, Z. Deng, C. Finn, M. Bansal, and H. Yao, “Analyzing and mitigating object hallucination in large vision-language models,” arXiv preprint arXiv:2310.00754 , 2023

    Pith/arXiv arXiv 2023

  3. [3]

    Evaluation and analysis of hallucination in large vision-language models,

    J. Wang, Y . Zhou, G. Xu, P. Shi, C. Zhao, H. Xu, Q. Ye, M. Yan, J. Zhang, J. Zhu et al. , “Evaluation and analysis of hallucination in large vision-language models,” arXiv preprint arXiv:2308.15126, 2023

    Pith/arXiv arXiv 2023

  4. [4]

    Calibrated self-rewarding vision language models,

    Y . Zhou, Z. Fan, D. Cheng, S. Yang, Z. Chen, C. Cui, X. Wang, Y . Li, L. Zhang, and H. Yao, “Calibrated self-rewarding vision language models,” Advances in Neural Information Processing Systems , vol. 37, pp. 51 503–51 531, 2024

    2024

  5. [5]

    Anyprefer: An agentic framework for preference data synthesis,

    Y . Zhou, Z. Wang, T. Wang, S. Xing, P. Xia, B. Li, K. Zheng, Z. Zhang, Z. Chen, W. Zheng et al. , “Anyprefer: An agentic framework for preference data synthesis,” arXiv preprint arXiv:2504.19276 , 2025

    Pith/arXiv arXiv 2025

  6. [6]

    Lumina-mgpt 2.0: Stand-alone autoregres- sive image modeling,

    Y . Xin, J. Yan, Q. Qin, Z. Li, D. Liu, S. Li, V . S.-J. Huang, Y . Zhou, R. Zhang, L. Zhuo et al., “Lumina-mgpt 2.0: Stand-alone autoregres- sive image modeling,” arXiv preprint arXiv:2507.17801 , 2025

    Pith/arXiv arXiv 2025

  7. [7]

    Parameter-efficient fine-tuning for pre-trained vision models: A sur- vey,

    Y . Xin, S. Luo, H. Zhou, J. Du, X. Liu, Y . Fan, Q. Li, and Y . Du, “Parameter-efficient fine-tuning for pre-trained vision models: A sur- vey,” arXiv preprint arXiv:2402.02242 , 2024

    arXiv 2024

  8. [8]

    V-petl bench: A unified visual parameter- efficient transfer learning benchmark,

    Y . Xin, S. Luo, X. Liu, H. Zhou, X. Cheng, C. E. Lee, J. Du, H. Wang, M. Chen, T. Liu et al. , “V-petl bench: A unified visual parameter- efficient transfer learning benchmark,” Advances in Neural Information Processing Systems, vol. 37, pp. 80 522–80 535, 2024

    2024

  9. [9]

    Vmt-adapter: Parameter- efficient transfer learning for multi-task dense scene understanding,

    Y . Xin, J. Du, Q. Wang, Z. Lin, and K. Yan, “Vmt-adapter: Parameter- efficient transfer learning for multi-task dense scene understanding,” in Proceedings of the AAAI Conference on Artificial Intelligence , vol. 38, no. 14, 2024, pp. 16 085–16 093

    2024

  10. [10]

    Mmap: Multi- modal alignment prompt for cross-domain multi-task learning,

    Y . Xin, J. Du, Q. Wang, K. Yan, and S. Ding, “Mmap: Multi- modal alignment prompt for cross-domain multi-task learning,” in Proceedings of the AAAI Conference on Artificial Intelligence , vol. 38, no. 14, 2024, pp. 16 076–16 084

    2024

  11. [11]

    Self-training with label- feature-consistency for domain adaptation,

    Y . Xin, S. Luo, P. Jin, Y . Du, and C. Wang, “Self-training with label- feature-consistency for domain adaptation,” in International Confer- ence on Database Systems for Advanced Applications. Springer, 2023, pp. 84–99

    2023

  12. [12]

    Lumina-image 2.0: A unified and efficient image generative framework,

    Q. Qin, L. Zhuo, Y . Xin, R. Du, Z. Li, B. Fu, Y . Lu, J. Yuan, X. Li, D. Liu et al., “Lumina-image 2.0: A unified and efficient image generative framework,” arXiv preprint arXiv:2503.21758 , 2025

    Pith/arXiv arXiv 2025

  13. [13]

    Towards understanding the working mechanism of text-to-image diffusion model,

    M. Yi, A. Li, Y . Xin, and Z. Li, “Towards understanding the working mechanism of text-to-image diffusion model,” Advances in Neural Information Processing Systems , vol. 37, pp. 55 342–55 369, 2024

    2024

  14. [14]

    Towards automated 3d evaluation of water leakage on a tunnel face via improved gan and self-attention dl model,

    C. Wu, H. Huang, L. Zhang, J. Chen, Y . Tong, and M. Zhou, “Towards automated 3d evaluation of water leakage on a tunnel face via improved gan and self-attention dl model,” Tunnelling and Underground Space Technology, vol. 142, p. 105432, 2023

    2023

  15. [15]

    Evaluation of tunnel rock mass integrity using multi-modal data and generative large model: Tunnel rip-gpt,

    C. Wu, H. Huang, and Y .-Q. Ni, “Evaluation of tunnel rock mass integrity using multi-modal data and generative large model: Tunnel rip-gpt,” Available at SSRN 5348429 , 2025

    2025

  16. [16]

    A novel tree- augmented bayesian network for predicting rock weathering degree using incomplete dataset,

    C. Wu, H. Huang, J. Chen, M. Zhou, and S. Han, “A novel tree- augmented bayesian network for predicting rock weathering degree using incomplete dataset,” International Journal of Rock Mechanics and Mining Sciences , vol. 183, p. 105933, 2024

    2024

  17. [17]

    Rock mass quality prediction on tunnel faces with incomplete multi-source dataset via tree-augmented naive bayesian network,

    H. Huang, C. Wu, M. Zhou, J. Chen, T. Han, and L. Zhang, “Rock mass quality prediction on tunnel faces with incomplete multi-source dataset via tree-augmented naive bayesian network,” International Journal of Mining Science and Technology , vol. 34, no. 3, pp. 323–337, 2024

    2024

  18. [18]

    Rankelectra: Semi-supervised pre-training of learning-to- rank electra for web-scale search,

    Y . Li, H. Xiong, Y . Zhang, J. Bian, T. Peng, X. Li, S. Wang, L. Kong, and D. Yin, “Rankelectra: Semi-supervised pre-training of learning-to- rank electra for web-scale search,” in Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V . 1 , 2025, pp. 2415–2425

    2025

  19. [19]

    M2oerank: Multi-objective mixture-of-experts enhanced ranking for satisfaction-oriented web search,

    Y . Li, H. Zhang, Y . Zhang, X. Ma, W. Ye, N. Song, S. Wang, H. Xiong, D. Yin, and L. Chen, “M2oerank: Multi-objective mixture-of-experts enhanced ranking for satisfaction-oriented web search,” in 2025 IEEE 41st International Conference on Data Engineering (ICDE) . IEEE Computer Society, 2025, pp. 4441–4454

    2025

  20. [20]

    Towards ai search paradigm,

    Y . Li, H. Cai, R. Kong, X. Chen, J. Chen, J. Yang, H. Zhang, J. Li, J. Wu, Y . Chen et al. , “Towards ai search paradigm,” arXiv preprint arXiv:2506.17188, 2025

    arXiv 2025

  21. [21]

    S2phere: Semi-supervised pre-training for web search over heteroge- neous learning to rank data,

    Y . Li, H. Xiong, L. Kong, Q. Wang, S. Wang, G. Chen, and D. Yin, “S2phere: Semi-supervised pre-training for web search over heteroge- neous learning to rank data,” in Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , 2023, pp. 4437–4448

    2023

  22. [22]

    Rankexpert: A mixture of textual-and-behavioral experts for multi-objective learning-to-rank in web search,

    Y . Li, H. Zhang, Y . Zhang, H. Cai, M. Cai, S. Wang, H. Xiong, D. Yin, and L. Chen, “Rankexpert: A mixture of textual-and-behavioral experts for multi-objective learning-to-rank in web search,” in Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V . 2, 2025, pp. 4437–4449

    2025

  23. [23]

    Coltr: Semi-supervised learning to rank with co-training and over-parameterization for web search,

    Y . Li, H. Xiong, Q. Wang, L. Kong, H. Liu, H. Li, J. Bian, S. Wang, G. Chen, D. Dou et al. , “Coltr: Semi-supervised learning to rank with co-training and over-parameterization for web search,” IEEE Transactions on Knowledge and Data Engineering , vol. 35, no. 12, pp. 12 542–12 555, 2023

    2023

  24. [24]

    Mhrr: Moocs recommender service with meta hierarchical reinforced ranking,

    Y . Li, H. Xiong, L. Kong, R. Zhang, F. Xu, G. Chen, and M. Li, “Mhrr: Moocs recommender service with meta hierarchical reinforced ranking,” IEEE Transactions on Services Computing , vol. 16, no. 6, pp. 4467–4480, 2023

    2023

  25. [25]

    Fultr: A large-scale fusion learning to rank dataset and its application for satisfaction-oriented ranking,

    Y . Li, H. Zhang, H. Zhang, H. Cai, X. Ma, S. Wang, H. Xiong, Z. Ren, M. de Rijke, and D. Yin, “Fultr: A large-scale fusion learning to rank dataset and its application for satisfaction-oriented ranking,” in Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V . 2, 2025, pp. 5583–5594

    2025

  26. [26]

    Rainy: Unlocking satellite calibration for deep learning in precipitation,

    Z. Yu, H. Chen, M. Y . I. Idris, and P. Wang, “Rainy: Unlocking satellite calibration for deep learning in precipitation,” arXiv preprint arXiv:2504.10776, 2025

    Pith/arXiv arXiv 2025

  27. [27]

    Satellitemaker: A diffusion-based framework for terrain-aware remote sensing image reconstruction,

    Z. Yu, M. Y . I. Idris, and P. Wang, “Satellitemaker: A diffusion-based framework for terrain-aware remote sensing image reconstruction,” arXiv preprint arXiv:2504.12112 , 2025

    Pith/arXiv arXiv 2025

  28. [28]

    Forgetme: Evaluating selective forgetting in generative models,

    ——, “Forgetme: Evaluating selective forgetting in generative models,” arXiv preprint arXiv:2504.12574 , 2025

    Pith/arXiv arXiv 2025

  29. [29]

    Satellitecalculator: A multi-task vision foundation model for quantitative remote sensing inversion,

    Z. Yu, M. Idris, and P. Wang, “Satellitecalculator: A multi-task vision foundation model for quantitative remote sensing inversion,” arXiv preprint arXiv:2504.13442, 2025

    Pith/arXiv arXiv 2025

  30. [30]

    Dancetext: Point-driven interactive text and image layer editing using diffusion models,

    Z. Yu, M. Y . I. Idris, P. Wang, and Y . Xia, “Dancetext: Point-driven interactive text and image layer editing using diffusion models,” arXiv preprint arXiv:2504.14108, 2025

    arXiv 2025

  31. [31]

    Dc4cr: When cloud removal meets diffusion control in remote sensing,

    Z. Yu, M. Y . I. Idris, and P. Wang, “Dc4cr: When cloud removal meets diffusion control in remote sensing,” arXiv preprint arXiv:2504.14785, 2025

    Pith/arXiv arXiv 2025

  32. [32]

    Satellitefor- mula: Multi-modal symbolic regression from remote sensing imagery for physics discovery,

    Z. Yu, M. Idris, P. Wang, Y . Xia, F. Ma, R. Qureshi et al., “Satellitefor- mula: Multi-modal symbolic regression from remote sensing imagery for physics discovery,” arXiv preprint arXiv:2506.06176 , 2025

    Pith/arXiv arXiv 2025

  33. [33]

    From physics to foundation models: A review of ai-driven quantitative remote sensing inversion,

    Z. Yu, M. Y . I. Idris, H. Wang, P. Wang, J. Chen, and K. Wang, “From physics to foundation models: A review of ai-driven quantitative remote sensing inversion,” arXiv preprint arXiv:2507.09081 , 2025

    Pith/arXiv arXiv 2025

  34. [34]

    Estimating forest carbon stock using enhanced resnet and sentinel-2 imagery

    J. Ren, L. Liu, Y . Wu, L. Ouyang, and Z. Yu, “Estimating forest carbon stock using enhanced resnet and sentinel-2 imagery.” Forests (19994907), vol. 16, no. 7, 2025

    2025

  35. [35]

    Reasoning in computer vi- sion: Taxonomy, models, tasks, and methodologies,

    A. Sarkar, M. Y . I. Idris, and Z. Yu, “Reasoning in computer vi- sion: Taxonomy, models, tasks, and methodologies,” arXiv preprint arXiv:2508.10523, 2025

    arXiv 2025

  36. [36]

    Ft2tf: First-person statement text-to-talking face generation,

    X. Diao, M. Cheng, W. Barrios, and S. Jin, “Ft2tf: First-person statement text-to-talking face generation,” in Proceedings of the Winter Conference on Applications of Computer Vision (WACV) , February 2025, pp. 4821–4830

    2025

  37. [37]

    Temporal working memory: Query-guided segment re- finement for enhanced multimodal understanding,

    X. Diao, C. Zhang, W. Wu, Z. Ouyang, P. Qing, M. Cheng, S. V osoughi, and J. Gui, “Temporal working memory: Query-guided segment re- finement for enhanced multimodal understanding,” arXiv preprint arXiv:2502.06020, 2025

    Pith/arXiv arXiv 2025

  38. [38]

    Learning sparsity for effective and efficient music performance question answer- ing,

    X. Diao, T. Yang, C. Zhang, W. Wu, M. Cheng, and J. Gui, “Learning sparsity for effective and efficient music performance question answer- ing,” arXiv preprint arXiv:2506.01319 , 2025

    Pith/arXiv arXiv 2025

  39. [39]

    Learning musical representations for music performance question an- swering,

    X. Diao, C. Zhang, T. Wu, M. Cheng, Z. Ouyang, W. Wu, and J. Gui, “Learning musical representations for music performance question an- swering,” in Findings of the Association for Computational Linguistics: EMNLP 2024, 2024

    2024

  40. [40]

    Soundmind: Rl-incentivized logic reasoning for audio-language models,

    X. Diao, C. Zhang, K. Kong, W. Wu, C. Ma, Z. Ouyang, P. Qing, S. V osoughi, and J. Gui, “Soundmind: Rl-incentivized logic reasoning for audio-language models,” arXiv preprint arXiv:2506.12935 , 2025

    arXiv 2025

  41. [41]

    Encoder: Entity mining and modification relation binding for composed image retrieval,

    Z. Li, Z. Chen, H. Wen, Z. Fu, Y . Hu, and W. Guan, “Encoder: Entity mining and modification relation binding for composed image retrieval,” in Proceedings of the AAAI Conference on Artificial Intelli- gence, vol. 39, no. 5, 2025, pp. 5101–5109

    2025

  42. [42]

    Finecir: Explicit parsing of fine-grained modification semantics for composed image retrieval,

    Z. Li, Z. Fu, Y . Hu, Z. Chen, H. Wen, and L. Nie, “Finecir: Explicit parsing of fine-grained modification semantics for composed image retrieval,” https://arxiv.org/abs/2503.21309, 2025

    Pith/arXiv arXiv 2025

  43. [43]

    Offset: Segmentation-based focus shift revision for composed image retrieval,

    Z. Chen, Y . Hu, Z. Li, Z. Fu, X. Song, and L. Nie, “Offset: Segmentation-based focus shift revision for composed image retrieval,”

    work page

  44. [44]

    Me- dian: Adaptive intermediate-grained aggregation network for composed image retrieval,

    Q. Huang, Z. Chen, Z. Li, C. Wang, X. Song, Y . Hu, and L. Nie, “Me- dian: Adaptive intermediate-grained aggregation network for composed image retrieval,” in Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing . IEEE, 2025, pp. 1–5

    work page 2025

  45. [45]

    Pair: Complementarity-guided disentanglement for composed image retrieval,

    Z. Fu, Z. Li, Z. Chen, C. Wang, X. Song, Y . Hu, and L. Nie, “Pair: Complementarity-guided disentanglement for composed image retrieval,” in Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing . IEEE, 2025, pp. 1–5

    work page 2025

  46. [46]

    TFB: Towards comprehensive and fair benchmarking of time series forecasting methods,

    X. Qiu, J. Hu, L. Zhou, X. Wu, J. Du, B. Zhang, C. Guo, A. Zhou, C. S. Jensen, Z. Sheng, and B. Yang, “TFB: Towards comprehensive and fair benchmarking of time series forecasting methods,” in Proc. VLDB Endow., 2024, pp. 2363–2377

    work page 2024

  47. [47]

    DUET: Dual clustering enhanced multivariate time series forecasting,

    X. Qiu, X. Wu, Y . Lin, C. Guo, J. Hu, and B. Yang, “DUET: Dual clustering enhanced multivariate time series forecasting,” in SIGKDD, 2025, pp. 1185–1196

    work page 2025

  48. [48]

    Tab: Unified benchmarking of time series anomaly detection methods,

    X. Qiu, Z. Li, W. Qiu, S. Hu, L. Zhou, X. Wu, Z. Li, C. Guo, A. Zhou, Z. Sheng, J. Hu, C. S. Jensen, and B. Yang, “Tab: Unified benchmarking of time series anomaly detection methods,” in Proc. VLDB Endow., 2025, pp. 2775–2789

    work page 2025

  49. [49]

    Rgp: Neural network pruning through regular graph with edges swap- ping,

    Z. Chen, J. Xiang, Y . Lu, Q. Xuan, Z. Wang, G. Chen, and X. Yang, “Rgp: Neural network pruning through regular graph with edges swap- ping,” IEEE Transactions on Neural Networks and Learning Systems , vol. 35, no. 10, pp. 14 671–14 683, 2023

    work page 2023

  50. [50]

    Understanding the dynamics of dnns using graph modularity,

    Y . Lu, W. Yang, Y . Zhang, Z. Chen, J. Chen, Q. Xuan, Z. Wang, and X. Yang, “Understanding the dynamics of dnns using graph modularity,” in European Conference on Computer Vision . Springer, 2022, pp. 225–242

    work page 2022

  51. [51]

    A generic layer pruning method for signal modulation recognition deep learning models,

    Y . Lu, Y . Zhu, Y . Li, D. Xu, Y . Lin, Q. Xuan, and X. Yang, “A generic layer pruning method for signal modulation recognition deep learning models,” IEEE Transactions on Cognitive Communications and Networking, 2024

    work page 2024

  52. [52]

    Reassessing layer pruning in llms: New insights and methods,

    Y . Lu, H. Cheng, Y . Fang, Z. Wang, J. Wei, D. Xu, Q. Xuan, X. Yang, and Z. Zhu, “Reassessing layer pruning in llms: New insights and methods,” arXiv preprint arXiv:2411.15558 , 2024

    Pith/arXiv arXiv 2024

  53. [53]

    Can pre-trained models assist in dataset distillation?

    Y . Lu, X. Chen, Y . Zhang, J. Gu, T. Zhang, Y . Zhang, X. Yang, Q. Xuan, K. Wang, and Y . You, “Can pre-trained models assist in dataset distillation?” arXiv preprint arXiv:2310.03295 , 2023

    Pith/arXiv arXiv 2023

  54. [54]

    From llm-anation to llm-orchestrator: Coordinating small models for data labeling,

    Y . Lu, Z. Ji, J. Du, Y . Shanqing, Q. Xuan, and T. Zhou, “From llm-anation to llm-orchestrator: Coordinating small models for data labeling,” arXiv preprint arXiv:2506.16393 , 2025

    Pith/arXiv arXiv 2025

  55. [55]

    Redtest: Towards measuring redundancy in deep neural networks effectively,

    Y . Lu, P. Zhang, J. Wang, L. Ma, X. Yang, and Q. Xuan, “Redtest: Towards measuring redundancy in deep neural networks effectively,” arXiv preprint arXiv:2411.10507 , 2024

    Pith/arXiv arXiv 2024

  56. [56]

    Sglp: A similarity guided fast layer partition pruning for compressing large deep models,

    Y . Li, Y . Lu, Z. Dong, C. Yang, Y . Chen, and J. Gou, “Sglp: A similarity guided fast layer partition pruning for compressing large deep models,” arXiv preprint arXiv:2410.14720 , 2024

    arXiv 2024

  57. [57]

    Sepprune: Structured pruning for efficient deep speech separation,

    Y . Li, K. Li, X. Yin, Z. Yang, J. Dong, Z. Dong, C. Yang, Y . Tian, and Y . Lu, “Sepprune: Structured pruning for efficient deep speech separation,” arXiv preprint arXiv:2505.12079 , 2025

    Pith/arXiv arXiv 2025

  58. [58]

    Graph-based similarity of neural network representations,

    Z. Chen, Y . Lu, J. Hu, W. Yang, Q. Xuan, Z. Wang, and X. Yang, “Graph-based similarity of neural network representations,” arXiv preprint arXiv:2111.11165, 2021

    Pith/arXiv arXiv 2021

  59. [59]

    Fcos: A two-stage recoverable model pruning framework for automatic modulation recognition,

    Y . Lu, T. Ma, Z. Wang, Z. Chen, D. Xu, Y . Lin, Q. Xuan, and G. Gui, “Fcos: A two-stage recoverable model pruning framework for automatic modulation recognition,” arXiv preprint arXiv:2505.21571 , 2025

    Pith/arXiv arXiv 2025

  60. [60]

    Duse: A data expansion framework for low-resource automatic modulation recognition based on active learning,

    Y . Lu, H. Gao, Z. Chen, D. Xu, Y . Lin, Q. Xuan, and G. Gui, “Duse: A data expansion framework for low-resource automatic modulation recognition based on active learning,”arXiv preprint arXiv:2507.12011, 2025

    Pith/arXiv arXiv 2025

  61. [61]

    Sr-init: An interpretable layer pruning method,

    H. Tang, Y . Lu, and Q. Xuan, “Sr-init: An interpretable layer pruning method,” in ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2023, pp. 1–5

    work page 2023

  62. [62]

    Frect: Frequency-augmented convolutional transformer for robust time series anomaly detection,

    W. Zhang, D. Xu, G. Yao, X. Lin, R. Guan, C. Du, R. Han, X. Xuan, and C. Luo, “Frect: Frequency-augmented convolutional transformer for robust time series anomaly detection,” in International Conference on Intelligent Computing . Springer, 2025, pp. 15–26

    work page 2025

  63. [63]

    A-mess: Anchor based multimodal embedding with semantic synchronization for multimodal intent recog- nition,

    Y . Shen, X. Lin, and W. Fan, “A-mess: Anchor based multimodal embedding with semantic synchronization for multimodal intent recog- nition,” arXiv preprint arXiv:2503.19474 , 2025

    Pith/arXiv arXiv 2025

  64. [64]

    Dual-channel heterophilic message passing for graph fraud detection,

    W. Zhang, J. Zhong, G. Yao, R. Han, X. Lin, Z. Zhang, and C. Luo, “Dual-channel heterophilic message passing for graph fraud detection,” arXiv preprint arXiv:2504.14205 , 2025

    Pith/arXiv arXiv 2025

  65. [65]

    Dconad: A differencing-based contrastive representation learning framework for time series anomaly detection,

    W. Zhang, X. Lin, W. Yu, G. Yao, Y . Li, R. Han, S. Xu, H. Shi, C. Luo et al., “Dconad: A differencing-based contrastive representation learning framework for time series anomaly detection,” arXiv preprint arXiv:2504.14204, 2025

    Pith/arXiv arXiv 2025

  66. [66]

    Com- bining population genomics and fitness qtls to identify the genetics of local adaptation in arabidopsis thaliana,

    N. Price, B. T. Moyers, L. Lopez, J. R. Lasky, J. G. Monroe, J. L. Mullen, C. G. Oakley, J. Lin, J. ˚Agren, D. R. Schrider et al. , “Com- bining population genomics and fitness qtls to identify the genetics of local adaptation in arabidopsis thaliana,” Proceedings of the National Academy of Sciences , vol. 115, no. 19, pp. 5028–5033, 2018

    work page 2018

  67. [67]

    Identification of polymorphisms associated with drought adaptation qtl in brassica napus by resequencing,

    R. S. Fletcher, D. Herrmann, J. L. Mullen, Q. Li, D. R. Schrider, N. Price, J. Lin, K. Grogan, A. Kern, and J. K. McKay, “Identification of polymorphisms associated with drought adaptation qtl in brassica napus by resequencing,” G3: Genes, Genomes, Genetics , vol. 6, no. 4, pp. 793–803, 2016

    work page 2016

  68. [68]

    Linking genomic signatures of selection to expression variation and direct evidence of local adaptation,

    N. Price, J. L. Mullen, J. Lin, C. Boucher, and J. K. McKay, “Linking genomic signatures of selection to expression variation and direct evidence of local adaptation,” bioRxiv, pp. 2020–08, 2020

    work page 2020

  69. [69]

    Cancer statistics,

    W. H. Organization, “Cancer statistics,” WHO Factsheets, 2022

    work page 2022

  70. [70]

    Epidemiology of hepatocellular carcinoma,

    H. B. El-Serag, “Epidemiology of hepatocellular carcinoma,” Gastroen- terology Clinics, vol. 48, no. 3, pp. 467–484, 2019

    work page 2019

  71. [71]

    Self-evolving agents with reflective and memory- augmented abilities,

    X. Liang, Y . He, Y . Xia, X. Song, J. Wang, M. Tao, L. Sun, X. Yuan, J. Su, K. Li et al., “Self-evolving agents with reflective and memory- augmented abilities,” arXiv preprint arXiv:2409.00872 , 2024

    Pith/arXiv arXiv 2024

  72. [72]

    Diagnosis of hepatocellular carci- noma: current guidelines and future directions,

    A. Sangiovanni and M. Colombo, “Diagnosis of hepatocellular carci- noma: current guidelines and future directions,” Hepatoma Research, 2020

    work page 2020

  73. [73]

    Cmat: A multi-agent collaboration tuning framework for enhancing small language models,

    X. Liang, Y . He, M. Tao, Y . Xia, J. Wang, T. Shi, J. Wang, and J. Yang, “Cmat: A multi-agent collaboration tuning framework for enhancing small language models,” arXiv preprint arXiv:2404.01663 , 2024

    Pith/arXiv arXiv 2024

  74. [74]

    Score: Story coherence and retrieval enhancement for ai narratives,

    Q. Yi, Y . He, J. Wang, X. Song, S. Qian, X. Yuan, L. Sun, Y . Xin, J. Tang, K. Li et al., “Score: Story coherence and retrieval enhancement for ai narratives,” arXiv preprint arXiv:2503.23512 , 2025

    arXiv 2025

  75. [75]

    Resurrect mask autoregressive modeling for efficient and scalable image generation,

    Y . Xin, L. Zhuo, Q. Qin, S. Luo, Y . Cao, B. Fu, Y . He, H. Li, G. Zhai, X. Liu et al., “Resurrect mask autoregressive modeling for efficient and scalable image generation,” arXiv preprint arXiv:2507.13032 , 2025

    Pith/arXiv arXiv 2025

  76. [76]

    Reagent-v: A reward-driven multi-agent framework for video understanding,

    Y . Zhou, Y . He, Y . Su, S. Han, J. Jang, G. Bertasius, M. Bansal, and H. Yao, “Reagent-v: A reward-driven multi-agent framework for video understanding,” arXiv preprint arXiv:2506.01300 , 2025

    Pith/arXiv arXiv 2025

  77. [77]

    Ddpm-moco: Advancing industrial surface defect generation and detection with generative and contrastive learning,

    Y . He, X. Wang, and T. Shi, “Ddpm-moco: Advancing industrial surface defect generation and detection with generative and contrastive learning,” in International Joint Conference on Artificial Intelligence . Springer, 2024, pp. 34–49

    work page 2024

  78. [78]

    Twin co-adaptive dialogue for progressive image generation,

    J. Wang, Y . He, Y . Zhong, X. Song, J. Su, Y . Feng, H. He, W. Zhu, X. Yuan, K. Luet al., “Twin co-adaptive dialogue for progressive image generation,” arXiv preprint arXiv:2504.14868 , 2025

    arXiv 2025

  79. [79]

    A survey on multimodal machine learning,

    T. Baltrusaitis, C. Ahuja, and L.-P. Morency, “A survey on multimodal machine learning,” IEEE Transactions on Pattern Analysis and Ma- chine Intelligence, 2019

    work page 2019

  80. [80]

    Multimodal deep learning: review, taxonomy, and open challenges,

    D. Ramachandram and G. W. Taylor, “Multimodal deep learning: review, taxonomy, and open challenges,” Information Fusion, 2017

    work page 2017

Showing first 80 references.