pith. machine review for the scientific record. sign in

arxiv: 2605.03820 · v1 · submitted 2026-05-05 · 💻 cs.CV · cs.LG· cs.MM

Recognition: unknown

Multimodal Learning on Low-Quality Data with Conformal Predictive Self-Calibration

Xun Jiang , Yufan Gu , Disen Hu , Yuqing Hou , Yazhou Yao , Fumin Shen , Heng Tao Shen , Xing Xu
Authors on Pith no claims yet

Pith reviewed 2026-05-07 17:49 UTC · model grok-4.3

classification 💻 cs.CV cs.LGcs.MM
keywords multimodal learningconformal predictionself-calibrationmodality imbalancenoisy labelsfeature fusiongradient reweightinglow-quality data
0
0 comments X

The pith

A conformal prediction loop lets multimodal models self-calibrate features and gradients on low-quality data.

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

The paper argues that modality imbalance and label noise in multimodal learning share a root cause in uncertainty about which modalities and instances are trustworthy during training. It proposes Conformal Predictive Self-Calibration (CPSC) as a unified training loop that uses a conformal predictor to decompose unimodal features, keep only the most robust components, and rescale gradients according to per-instance reliability scores. The conformal predictor itself is updated on the fly so the whole system evolves together without any separate clean validation set. Experiments across six benchmarks show consistent gains over prior state-of-the-art methods when both imbalance and noise are present. The central promise is therefore a single mechanism that simultaneously strengthens representation resilience and steers optimization away from unreliable signals.

Core claim

Conformal Predictive Self-Calibration equips a multimodal model with an on-the-fly self-calibrating loop: a conformal predictor decomposes each unimodal feature into components and selects the robust subset for fusion (Representation Self-Calibration), while also computing instance-wise reliability scores that reweight gradients during back-propagation (Gradient Self-Calibration); the predictor is itself updated from the model’s evolving predictions so that calibration and learning co-evolve without external clean data.

What carries the argument

The self-calibrating training loop that interleaves a conformal predictor’s robustness scoring with feature decomposition and gradient reweighting.

If this is right

  • The same loop can be applied to any multimodal architecture without changing the backbone or loss.
  • Separate techniques for imbalance correction and noise robustness can be replaced by one mechanism.
  • Training no longer requires a held-out clean validation set to estimate modality or instance reliability.
  • The framework produces per-instance reliability scores that can be inspected or used for downstream filtering.
  • Performance remains stable when the degree of imbalance or noise changes during training.

Where Pith is reading between the lines

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

  • The approach may generalize to other uncertainty sources such as missing modalities or domain shift if the conformal scoring is kept unchanged.
  • Because the calibration runs continuously, the method could be extended to online or lifelong multimodal settings where data quality drifts over time.
  • If the conformal predictor’s coverage guarantees hold under distribution shift, the reliability scores could serve as uncertainty estimates for active learning or human-in-the-loop correction.

Load-bearing premise

A conformal predictor can reliably identify robust feature components and instance-wise reliability scores from the model’s own evolving predictions without introducing bias or requiring separate clean validation data.

What would settle it

Ablation experiments in which the conformal predictor is replaced by a fixed threshold or random selection show no performance gain over standard multimodal training on the same imbalanced-plus-noisy datasets.

Figures

Figures reproduced from arXiv: 2605.03820 by Disen Hu, Fumin Shen, Heng Tao Shen, Xing Xu, Xun Jiang, Yazhou Yao, Yufan Gu, Yuqing Hou.

Figure 1
Figure 1. Figure 1: Implicit imbalance and explicit noise corruption would view at source ↗
Figure 2
Figure 2. Figure 2: The overall architecture of our CPSC framework, illustrating the self-calibration training loop with Representation Self view at source ↗
Figure 3
Figure 3. Figure 3: Analysis of the RSC module on the CREMA-D (Left) view at source ↗
Figure 4
Figure 4. Figure 4: Analysis on the conformal predictor updating frequency view at source ↗
Figure 6
Figure 6. Figure 6: Analysis of reliability of initialized and trained models view at source ↗
Figure 5
Figure 5. Figure 5: Comparative visualizations of audio and visual feature view at source ↗
read the original abstract

Multimodal learning often grapples with the challenge of low-quality data, which predominantly manifests as two facets: modality imbalance and noisy corruption. While these issues are often studied in isolation, we argue that they share a common root in the predictive uncertainty towards the reliability of individual modalities and instances during learning. In this paper, we propose a unified framework, termed Conformal Predictive Self-Calibration (CPSC), which leverages conformal prediction to equip the model with the ability to perform self-guided calibration on-the-fly. The core of our proposed CPSC lies in a novel self-calibrating training loop that seamlessly integrates two key modules: (1) Representation Self-Calibration, which decomposes unimodal features into components, and selectively fuses the most robust ones identified by a conformal predictor to enhance feature resilience. (2) Gradient Self-Calibration, which recalibrates the gradient flow during backpropagation based on instance-wise reliability scores, steering the optimization towards more trustworthy directions. Furthermore, we also devise a self-update strategy for the conformal predictor to ensure the entire system co-evolves consistently throughout the training process. Extensive experiments on six benchmark datasets under both imbalanced and noisy settings demonstrate that our CPSC framework consistently outperforms existing state-of-the-art methods. Our code is available at https://github.com/XunCHN/CPSC.

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 proposes Conformal Predictive Self-Calibration (CPSC), a unified framework for multimodal learning on low-quality data characterized by modality imbalance and label noise. It introduces a self-calibrating training loop that integrates Representation Self-Calibration (decomposing unimodal features and selectively fusing robust components via a conformal predictor) and Gradient Self-Calibration (recalibrating gradients using instance-wise reliability scores), together with a self-update strategy for the conformal predictor to enable co-evolution during training. The central claim is that this approach consistently outperforms existing state-of-the-art methods across six benchmark datasets under both imbalanced and noisy settings.

Significance. If the self-calibration loop can be rigorously shown to improve rather than degrade performance without bias amplification, the work would offer a principled, on-the-fly mechanism for robustness in multimodal models that avoids reliance on separate clean validation data. The public release of code at the provided GitHub link is a clear strength supporting reproducibility.

major comments (2)
  1. [Abstract and Method] Abstract and Method section: The self-update strategy for the conformal predictor is described as relying on the model's own evolving predictions to identify robust feature components and instance-wise reliability scores. This setup risks circular dependency, since standard conformal prediction requires exchangeable calibration data independent of the model being calibrated; the manuscript provides no theoretical analysis or early-training ablation demonstrating that initial unreliable predictions (dominated by stronger modalities or noise) do not systematically bias the calibration and amplify errors rather than mitigate them.
  2. [Experiments] Experiments section: The claim of consistent outperformance on six benchmarks under imbalanced and noisy settings lacks reported details on statistical significance testing, ablation studies isolating the conformal predictor's contribution, or sensitivity analysis to noise/imbalance levels. Without these, it is impossible to confirm that gains are attributable to the proposed self-calibration rather than other factors.
minor comments (2)
  1. [Abstract] Abstract: Implementation specifics for the conformal predictor (e.g., nonconformity score definition, how the self-update is performed without external data) are absent, hindering immediate assessment of novelty and reproducibility.
  2. [Overall] Overall: Notation for reliability scores and feature decomposition components could be clarified with explicit equations or pseudocode to improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thoughtful and constructive comments. We address each major point below and describe the revisions planned for the next version of the manuscript.

read point-by-point responses

  1. Referee: [Abstract and Method] Abstract and Method section: The self-update strategy for the conformal predictor is described as relying on the model's own evolving predictions to identify robust feature components and instance-wise reliability scores. This setup risks circular dependency, since standard conformal prediction requires exchangeable calibration data independent of the model being calibrated; the manuscript provides no theoretical analysis or early-training ablation demonstrating that initial unreliable predictions (dominated by stronger modalities or noise) do not systematically bias the calibration and amplify errors rather than mitigate them.

    Authors: We acknowledge the referee's concern about potential circular dependency. The self-update strategy is intended to allow the conformal predictor to co-evolve with the model by progressively incorporating more reliable samples as training proceeds, rather than relying on a fixed independent calibration set. While the manuscript does not contain a formal theoretical proof of non-amplification, the empirical results across multiple noisy and imbalanced settings indicate that performance improves rather than degrades. To address the comment directly, we will add an early-training ablation study that tracks calibration quality, prediction set sizes, and downstream accuracy from epoch 1 onward, together with a short discussion of the conservative sample-selection heuristic used to initialize the predictor. revision: yes

  2. Referee: [Experiments] Experiments section: The claim of consistent outperformance on six benchmarks under imbalanced and noisy settings lacks reported details on statistical significance testing, ablation studies isolating the conformal predictor's contribution, or sensitivity analysis to noise/imbalance levels. Without these, it is impossible to confirm that gains are attributable to the proposed self-calibration rather than other factors.

    Authors: We agree that these elements would strengthen the experimental claims. In the revised manuscript we will (i) report statistical significance (paired t-tests or Wilcoxon signed-rank tests over five random seeds) for all main-table comparisons, (ii) expand the ablation section with variants that disable the conformal predictor while keeping other components fixed, and (iii) add sensitivity plots that vary the degree of modality imbalance and label-noise ratio on at least two representative datasets. These additions will make the attribution of gains to the self-calibration components explicit. revision: yes

Circularity Check

0 steps flagged

No significant circularity in the CPSC self-calibration framework

full rationale

The paper describes an iterative self-calibrating training loop that integrates conformal prediction for representation and gradient recalibration, along with a self-update strategy for the predictor to allow co-evolution. No equations or definitions are provided in the available text that reduce the reliability scores, robust feature components, or final predictions to the inputs by construction (e.g., no self-definitional mapping where the conformal output is fitted directly from the target quantity it is claimed to predict). The framework is presented as a procedural training mechanism evaluated on external benchmark datasets under imbalanced and noisy conditions, rendering the central claims self-contained rather than tautological. No load-bearing self-citation chains or ansatz smuggling are exhibited.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract provides insufficient detail to enumerate specific free parameters or invented entities; the framework implicitly relies on standard conformal prediction validity assumptions.

axioms (1)
  • domain assumption Conformal prediction yields valid uncertainty estimates that can be used to select robust features and reweight gradients during training
    Central to both Representation Self-Calibration and Gradient Self-Calibration modules.

pith-pipeline@v0.9.0 · 5561 in / 1177 out tokens · 28846 ms · 2026-05-07T17:49:43.200503+00:00 · methodology

Review history (3 revisions) →

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

60 extracted references · 4 canonical work pages · 2 internal anchors

  1. [1]

    Indoor segmentation and support inference from rgbd images

    Nathan Silberman, Derek Hoiem, Pushmeet Kohli, and Rob Fergus. Indoor segmentation and support inference from rgbd images. InECCV, pages 746–760. Springer, 2012. 1, 5

    2012

  2. [2]

    Crema-d: Crowd-sourced emotional multimodal actors dataset.IEEE TAFFC, 5(4):377–390, 2014

    Houwei Cao, David G Cooper, Michael K Keutmann, Ruben C Gur, Ani Nenkova, and Ragini Verma. Crema-d: Crowd-sourced emotional multimodal actors dataset.IEEE TAFFC, 5(4):377–390, 2014. 1, 5

    2014

  3. [3]

    Empowering visible- infrared person re-identification with large foundation mod- els.NeurIPS, 37:117363–117387, 2024

    Zhangyi Hu, Bin Yang, and Mang Ye. Empowering visible- infrared person re-identification with large foundation mod- els.NeurIPS, 37:117363–117387, 2024. 1, 5, 6, 8

    2024

  4. [4]

    A survey of mul- timodal learning: Methods, applications, and future.ACM Comput

    Yuan Yuan, Zhaojian Li, and Bin Zhao. A survey of mul- timodal learning: Methods, applications, and future.ACM Comput. Surv., 57(7):1–34, 2025. 1

    2025

  5. [5]

    Foundations & trends in multimodal machine learning: Prin- ciples, challenges, and open questions.ACM Comput

    Paul Pu Liang, Amir Zadeh, and Louis-Philippe Morency. Foundations & trends in multimodal machine learning: Prin- ciples, challenges, and open questions.ACM Comput. Surv., 56(10):1–42, 2024

    2024

  6. [6]

    Joint objective and subjective fuzziness denoising for multimodal sentiment analysis.IEEE TFS, 33(1):15–27, 2024

    Xun Jiang, Xing Xu, Huimin Lu, Lianghua He, and Heng Tao Shen. Joint objective and subjective fuzziness denoising for multimodal sentiment analysis.IEEE TFS, 33(1):15–27, 2024. 1, 2

    2024

  7. [7]

    Multimodal fusion on low-quality data: A comprehensive survey.arXiv preprint arXiv:2404.18947, 2024

    Qingyang Zhang, Yake Wei, Zongbo Han, Huazhu Fu, Xi Peng, Cheng Deng, Qinghua Hu, Cai Xu, Jie Wen, Di Hu, et al. Multimodal fusion on low-quality data: A comprehen- sive survey.arXiv preprint arXiv:2404.18947, 2024. 1

    work page arXiv 2024

  8. [8]

    Embracing unimodal aleatoric uncertainty for robust multimodal fusion

    Zixian Gao, Xun Jiang, Xing Xu, Fumin Shen, Yujie Li, and Heng Tao Shen. Embracing unimodal aleatoric uncertainty for robust multimodal fusion. InCVPR, pages 26876–26885,

  9. [9]

    Balanced multimodal learning via on-the-fly gradient modulation

    Xiaokang Peng, Yake Wei, Andong Deng, Dong Wang, and Di Hu. Balanced multimodal learning via on-the-fly gradient modulation. InCVPR, pages 8238–8247, 2022. 1, 2

    2022

  10. [10]

    Mmpareto: Boosting multimodal learning with innocent unimodal assistance.ICML, 2024

    Yake Wei and Di Hu. Mmpareto: Boosting multimodal learning with innocent unimodal assistance.ICML, 2024. 1

    2024

  11. [11]

    Provable dynamic fusion for low-quality multimodal data

    Qingyang Zhang, Haitao Wu, Changqing Zhang, Qinghua Hu, Huazhu Fu, Joey Tianyi Zhou, and Xi Peng. Provable dynamic fusion for low-quality multimodal data. InICML, pages 41753–41769, 2023. 1, 2, 5

    2023

  12. [12]

    Intra-and inter-modal curriculum for multi- modal learning

    Yuwei Zhou, Xin Wang, Hong Chen, Xuguang Duan, and Wenwu Zhu. Intra-and inter-modal curriculum for multi- modal learning. InACM MM, pages 3724–3735, 2023. 1

    2023

  13. [13]

    Geometric gradient di- vergence modulation for imbalanced multimodal learning

    Disen Hu, Xun Jiang, Zhe Sun, Hao Yang, Chong Peng, Peng Yan, Heng Tao Shen, and Xing Xu. Geometric gradient di- vergence modulation for imbalanced multimodal learning. In ACM MM, pages 1337–1345, 2025. 1, 2

    2025

  14. [14]

    Towards balanced active learning for multimodal classification

    Meng Shen, Yizheng Huang, Jianxiong Yin, Heqing Zou, Deepu Rajan, and Simon See. Towards balanced active learning for multimodal classification. InACM MM, pages 3434–3445, 2023. 1

    2023

  15. [15]

    Improving mul- timodal learning balance and sufficiency through data remix- ing.ICML, 2025

    Xiaoyu Ma, Hao Chen, and Yongjian Deng. Improving mul- timodal learning balance and sufficiency through data remix- ing.ICML, 2025. 1

    2025

  16. [16]

    Ro- bust multimodal learning via representation decoupling

    Shicai Wei, Yang Luo, Yuji Wang, and Chunbo Luo. Ro- bust multimodal learning via representation decoupling. In ECCV, pages 38–54. Springer, 2024. 2

    2024

  17. [17]

    A tutorial on conformal prediction.JMLR, 9(3), 2008

    Glenn Shafer and Vladimir V ovk. A tutorial on conformal prediction.JMLR, 9(3), 2008. 2, 3

    2008

  18. [18]

    A simple baseline for bayesian uncertainty in deep learning.NeurIPS, 32, 2019

    Wesley J Maddox, Pavel Izmailov, Timur Garipov, Dmitry P Vetrov, and Andrew Gordon Wilson. A simple baseline for bayesian uncertainty in deep learning.NeurIPS, 32, 2019. 2

    2019

  19. [19]

    A dynamic bayesian network based framework for multimodal context- aware interactions

    Violet Yinuo Han, Tianyi Wang, Hyunsung Cho, Kashyap Todi, Ajoy Savio Fernandes, Andre Levi, Zheng Zhang, Tovi Grossman, Alexandra Ion, and Tanya R Jonker. A dynamic bayesian network based framework for multimodal context- aware interactions. InIUI, pages 54–69, 2025. 2

    2025

  20. [20]

    Zero-shot video moment retrieval with angular reconstructive text embeddings.IEEE TMM, 26:9657–9670, 2024

    Xun Jiang, Xing Xu, Zailei Zhou, Yang Yang, Fumin Shen, and Heng Tao Shen. Zero-shot video moment retrieval with angular reconstructive text embeddings.IEEE TMM, 26:9657–9670, 2024. 2

    2024

  21. [21]

    Evidence-based multi- feature fusion for adversarial robustness.IEEE TPAMI, 2025

    Zheng Wang, Xing Xu, Lei Zhu, Yi Bin, Guoqing Wang, Yang Yang, and Heng Tao Shen. Evidence-based multi- feature fusion for adversarial robustness.IEEE TPAMI, 2025

    2025

  22. [22]

    Generalizable egocentric task ver- ification via cross-modal hybrid hypergraph matching.IEEE TPAMI, 2026

    Xun Jiang, Xing Xu, Zheng Wang, Jingkuan Song, Fumin Shen, and Heng Tao Shen. Generalizable egocentric task ver- ification via cross-modal hybrid hypergraph matching.IEEE TPAMI, 2026. 2

    2026

  23. [23]

    Multi- set feature learning for highly imbalanced data classification

    Xiao-Yuan Jing, Xinyu Zhang, Xiaoke Zhu, Fei Wu, Xinge You, Yang Gao, Shiguang Shan, and Jing-Yu Yang. Multi- set feature learning for highly imbalanced data classification. IEEE TPAMI, 43(1):139–156, 2019. 2

    2019

  24. [24]

    Learning with privileged multi- modal knowledge for unimodal segmentation.IEEE TMI, 41(3):621–632, 2021

    Cheng Chen, Qi Dou, Yueming Jin, Quande Liu, and Pheng Ann Heng. Learning with privileged multi- modal knowledge for unimodal segmentation.IEEE TMI, 41(3):621–632, 2021. 2

    2021

  25. [25]

    Adaptive unimodal regulation for balanced multimodal in- formation acquisition

    Chengxiang Huang, Yake Wei, Zequn Yang, and Di Hu. Adaptive unimodal regulation for balanced multimodal in- formation acquisition. InCVPR, pages 25854–25863, 2025. 2, 6

    2025

  26. [26]

    Mmpareto: Boosting multimodal learning with innocent unimodal assistance.ICML, 2024

    Yake Wei and Di Hu. Mmpareto: Boosting multimodal learning with innocent unimodal assistance.ICML, 2024. 2, 5, 6

    2024

  27. [27]

    Trusted multi-view classification with dynamic evi- dential fusion.IEEE TPAMI, 45(2):2551–2566, 2022

    Zongbo Han, Changqing Zhang, Huazhu Fu, and Joey Tianyi Zhou. Trusted multi-view classification with dynamic evi- dential fusion.IEEE TPAMI, 45(2):2551–2566, 2022. 2

    2022

  28. [28]

    Dynamic evidence decoupling for trusted multi-view learning

    Ying Liu, Lihong Liu, Cai Xu, Xiangyu Song, Ziyu Guan, and Wei Zhao. Dynamic evidence decoupling for trusted multi-view learning. InACM MM, pages 7269–7277, 2024. 2

    2024

  29. [29]

    Resisting noise in pseudo la- bels: Audible video event parsing with evidential learning

    Xun Jiang, Xing Xu, Liqing Zhu, Zhe Sun, Andrzej Ci- chocki, and Heng Tao Shen. Resisting noise in pseudo la- bels: Audible video event parsing with evidential learning. IEEE TNNLS, 36(6):10874–10888, 2024. 2

    2024

  30. [30]

    Cross-modal uncertainty modeling with diffusion-based refinement for text-based person retrieval

    Shenshen Li, Xing Xu, Chen He, Fumin Shen, Yang Yang, and Heng Tao Shen. Cross-modal uncertainty modeling with diffusion-based refinement for text-based person retrieval. IEEE TCSVT, 35(3):2881–2893, 2024. 2

    2024

  31. [31]

    Adaptive uncertainty-based learning for text-based person retrieval

    Shenshen Li, Chen He, Xing Xu, Fumin Shen, Yang Yang, and Heng Tao Shen. Adaptive uncertainty-based learning for text-based person retrieval. InAAAI, volume 38, pages 3172–3180, 2024. 2

    2024

  32. [32]

    Ensemble quantile networks: Uncertainty-aware reinforcement learn- ing with applications in autonomous driving.IEEE TITS, 24(6):6030–6041, 2023

    Carl-Johan Hoel, Krister Wolff, and Leo Laine. Ensemble quantile networks: Uncertainty-aware reinforcement learn- ing with applications in autonomous driving.IEEE TITS, 24(6):6030–6041, 2023. 2

    2023

  33. [33]

    Uncertainty quantifi- cation of collaborative detection for self-driving

    Sanbao Su, Yiming Li, Sihong He, Songyang Han, Chen Feng, Caiwen Ding, and Fei Miao. Uncertainty quantifi- cation of collaborative detection for self-driving. InICRA, 2023

    2023

  34. [34]

    Hyperdi- mensional uncertainty quantification for multimodal uncer- tainty fusion in autonomous vehicles perception

    Luke Chen, Junyao Wang, Trier Mortlock, Pramod Khar- gonekar, and Mohammad Abdullah Al Faruque. Hyperdi- mensional uncertainty quantification for multimodal uncer- tainty fusion in autonomous vehicles perception. InCVPR, pages 22306–22316, 2025. 2

    2025

  35. [35]

    Hyper-opinion vague- ness quantification for robust multimodal learning

    Disen Hu, Xun Jiang, Xiaofeng Cao, Zheng Wang, Jingkuan Song, Heng Tao Shen, and Xing Xu. Hyper-opinion vague- ness quantification for robust multimodal learning. InAAAI, volume 40, pages 21831–21839, 2026

    2026

  36. [36]

    Uncertainty-debiased multimodal fusion: Learning deterministic joint representation for multimodal sentiment analysis

    Zixian Gao, Xun Jiang, Hua Chen, Yujie Li, Yang Yang, and Xing Xu. Uncertainty-debiased multimodal fusion: Learning deterministic joint representation for multimodal sentiment analysis. InICME, pages 1–6, 2024. 2

    2024

  37. [37]

    Aligning model properties via conformal risk control

    William Overman, Jacqueline Vallon, and Mohsen Bay- ati. Aligning model properties via conformal risk control. NeurIPS, 37:110702–110722, 2024. 2

    2024

  38. [38]

    Uncertainty- aware real-time visual anomaly detection with conformal prediction in dynamic indoor environments.IEEE RAL,

    Arya Saboury and Mustafa Kemal Uyguroglu. Uncertainty- aware real-time visual anomaly detection with conformal prediction in dynamic indoor environments.IEEE RAL,

  39. [39]

    Confor- mal graph-level out-of-distribution detection with adaptive data augmentation

    Xixun Lin, Yanan Cao, Nan Sun, Lixin Zou, Chuan Zhou, Peng Zhang, Shuai Zhang, Ge Zhang, and Jia Wu. Confor- mal graph-level out-of-distribution detection with adaptive data augmentation. InWWW, pages 4755–4765, 2025. 2

    2025

  40. [40]

    Analyzing uncertainty of llm-as-a-judge: Interval evaluations with conformal prediction.EMNLP, 2025

    Huanxin Sheng, Xinyi Liu, Hangfeng He, Jieyu Zhao, and Jian Kang. Analyzing uncertainty of llm-as-a-judge: Interval evaluations with conformal prediction.EMNLP, 2025. 2

    2025

  41. [41]

    Conformal alignment: Knowing when to trust foundation models with guarantees

    Yu Gui, Ying Jin, and Zhimei Ren. Conformal alignment: Knowing when to trust foundation models with guarantees. NeurIPS, 37:73884–73919, 2024. 2

    2024

  42. [42]

    Con- formal prediction with learned features

    Shayan Kiyani, George J Pappas, and Hamed Hassani. Con- formal prediction with learned features. InICML, pages 24749–24769, 2024. 3

    2024

  43. [43]

    Any2any: Incomplete multi- modal retrieval with conformal prediction.arXiv preprint arXiv:2411.10513, 2024

    Po-han Li, Yunhao Yang, Mohammad Omama, Sandeep Chinchali, and Ufuk Topcu. Any2any: Incomplete multi- modal retrieval with conformal prediction.arXiv preprint arXiv:2411.10513, 2024

    work page arXiv 2024

  44. [44]

    Confor- mal prediction for multimodal regression.arXiv preprint arXiv:2410.19653, 2024

    Alexis Bose, Jonathan Ethier, and Paul Guinand. Confor- mal prediction for multimodal regression.arXiv preprint arXiv:2410.19653, 2024. 3

    work page internal anchor Pith review arXiv 2024

  45. [45]

    Mutual information- calibrated conformal feature fusion for uncertainty-aware multimodal 3d object detection at the edge

    Alex C Stutts, Danilo Erricolo, Sathya Ravi, Theja Tula- bandhula, and Amit Ranjan Trivedi. Mutual information- calibrated conformal feature fusion for uncertainty-aware multimodal 3d object detection at the edge. InICRA, pages 2029–2035, 2024. 3

    2029

  46. [46]

    Conformal- ized multimodal uncertainty regression and reasoning

    Domenico Parente, Nastaran Darabi, Alex C Stutts, Theja Tulabandhula, and Amit Ranjan Trivedi. Conformal- ized multimodal uncertainty regression and reasoning. In ICASSP, pages 6985–6989, 2024

    2024

  47. [47]

    Multi-model on- line conformal prediction with graph-structured feedback

    Erfan Hajihashemi and Yanning Shen. Multi-model on- line conformal prediction with graph-structured feedback. TMLR. 3

  48. [48]

    Reconboost: Boosting can achieve modal- ity reconcilement.ICML, 2024

    Cong Hua, Qianqian Xu, Shilong Bao, Zhiyong Yang, and Qingming Huang. Reconboost: Boosting can achieve modal- ity reconcilement.ICML, 2024. 6

    2024

  49. [49]

    Towards equi- librium: An instantaneous probe-and-rebalance multimodal learning approach

    Yang Yang, Xixian Wu, and Qing-Yuan Jiang. Towards equi- librium: An instantaneous probe-and-rebalance multimodal learning approach. InIJCAI, pages 3552–3560, 2025. 6

    2025

  50. [50]

    Improving mul- timodal learning via imbalanced learning

    Shicai Wei, Chunbo Luo, and Yang Luo. Improving mul- timodal learning via imbalanced learning. InICCV, pages 2250–2259, 2025. 6

    2025

  51. [51]

    Boosting multi- modal learning via disentangled gradient learning

    Shicai Wei, Chunbo Luo, and Yang Luo. Boosting multi- modal learning via disentangled gradient learning. InICCV, pages 22879–22888, 2025. 6

    2025

  52. [52]

    Audio-visual event localization in unconstrained videos

    Yapeng Tian, Jing Shi, Bochen Li, Zhiyao Duan, and Chen- liang Xu. Audio-visual event localization in unconstrained videos. InECCV, pages 247–263, 2018. 5

    2018

  53. [53]

    The Kinetics Human Action Video Dataset

    Will Kay, Joao Carreira, Karen Simonyan, Brian Zhang, Chloe Hillier, Sudheendra Vijayanarasimhan, Fabio Viola, Tim Green, Trevor Back, Paul Natsev, et al. The kinetics hu- man action video dataset.arXiv preprint arXiv:1705.06950,

    work page internal anchor Pith review arXiv

  54. [54]

    Sun rgb-d: A rgb-d scene understanding benchmark suite

    Shuran Song, Samuel P Lichtenberg, and Jianxiong Xiao. Sun rgb-d: A rgb-d scene understanding benchmark suite. In CVPR, pages 567–576, 2015. 5

    2015

  55. [55]

    Sentiment analysis on multi-view social data

    Teng Niu, Shiai Zhu, Lei Pang, and Abdulmotaleb El Sad- dik. Sentiment analysis on multi-view social data. InMMM, pages 15–27. Springer, 2016. 5

    2016

  56. [56]

    Reliable conflictive multi-view learning

    Cai Xu, Jiajun Si, Ziyu Guan, Wei Zhao, Yue Wu, and Xiyue Gao. Reliable conflictive multi-view learning. InAAAI, vol- ume 38, pages 16129–16137, 2024. 6

    2024

  57. [57]

    Noisy label calibra- tion for multi-view classification

    Shilin Xu, Yuan Sun, Xingfeng Li, Siyuan Duan, Zhenwen Ren, Zheng Liu, and Dezhong Peng. Noisy label calibra- tion for multi-view classification. InAAAI, volume 39, pages 21797–21805, 2025. 6

    2025

  58. [58]

    Stochastic gradient descent tricks

    L ´eon Bottou. Stochastic gradient descent tricks. InNeural networks: tricks of the trade: second edition, pages 421–436

  59. [59]

    A method for stochastic opti- mization.ICLR, 1412(6), 2014

    Kingma DP Ba J Adam et al. A method for stochastic opti- mization.ICLR, 1412(6), 2014. 7

    2014

  60. [60]

    Adaptive sub- gradient methods for online learning and stochastic opti- mization.JMLR, 12(7), 2011

    John Duchi, Elad Hazan, and Yoram Singer. Adaptive sub- gradient methods for online learning and stochastic opti- mization.JMLR, 12(7), 2011. 7

    2011