Pith. sign in

REVIEW 5 major objections 6 minor 76 references

Fusing Cross-Domain Knowledge from Multimodal Data to Solve Problems in the Physical World

T0 review · 5 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read The paper claims that cross-domain multimodal data fusion is a distinct problem needing its own methodology, and it supplies a four-layer framework for choosing, linking, fusing, and transforming data across domains.

desk verdict A useful organizing framework for cross-domain multimodal fusion, but the effectiveness claim is not supported and the link-identification step remains unvalidated. read the letter →

arxiv 2506.03155 v2 pith:CIE24RMZ submitted 2025-05-20 cs.LG cs.AI

classification cs.LGcs.AI
keywords knowledgefusionmultimodaldatacross-domaintransformationurbancomputingspatio-temporalalignmentdeeplearning
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

This paper argues that existing multimodal data fusion methods assume the datasets are intrinsically aligned, which fails when data comes from different domains and must be combined to solve real-world problems. The author formally defines cross-domain knowledge fusion and proposes a four-layer framework that answers what to fuse, why fusion is possible, and how to fuse. The paper identifies knowledge alignment as the core challenge and offers four principles to justify links between disparate datasets. A sympathetic reader would see this as a systematic blueprint for turning available data from multiple domains into solutions for physical-world tasks without collecting new data.

What carries the argument

The key machinery is the four-layer framework together with the philosophy of knowledge alignment. The framework's load-bearing part is the Links Layer, which proposes four alignment principles: multiview-based (different views of the same object collectively form a better representation), similarity-based (similar objects complement each other's sparse data), dependency-based (probabilistic dependencies between objects constrain and enrich estimation), and commonality-based (shared structure between domains allows knowledge transfer). These principles justify why disparate datasets can be fused and guide the design of model structures, while the two fusion paradigms (precise and coarse) and data-transformation components handle the how.

What would settle it

A concrete test would be to take a task where the framework's procedure identifies a purported causal link between two domains, then construct a holdout dataset where that link is actually absent or reversed; if the fused model performs worse than the best single-domain baseline on that dataset, the framework's link-discovery step lacks the validation it needs. A simpler calculation: on a series of urban tasks (e.g., air-quality inference, traffic flow prediction), measure whether the performance gain from cross-domain fusion correlates with the strength of the statistical link between the contributing datasets; a near-zero or negative correlation would refute the central claim that domain-knowledge-derived links reliably justify fusion.

Watch

Extended reading notes

Core claim

The central claim is that cross-domain multimodal data fusion is a new research theme distinct from single-domain data fusion, and that it can be operationalized through a four-layer framework consisting of Domains, Links, Models, and Data layers. The Domains Layer selects relevant data by tracing a problem backwards through root causes and contributing factors. The Links Layer supplies four principles of knowledge alignment (multiview-based, similarity-based, dependency-based, and commonality-based) to explain why data from different domains can be fused. The Models Layer distinguishes two paradigms, precise fusion and coarse fusion, and the Data Layer transforms different structures, resolutions, scales, and distributions into a consistent representation. The paper claims that following these layers and the ten associated procedures yields effective fusion for real-world problems.

Load-bearing premise

The framework assumes that meaningful, accurate links between datasets from different domains can be discovered from domain knowledge and the four alignment principles, and that these links are correct enough that fusing the data improves rather than harms the result.

Editorial extensions

If this is right

  • If the framework is correct, practitioners can systematically decide whether to fuse data from other domains before building a model, instead of relying on data availability or ad hoc intuition.
  • The four alignment principles could serve as a checklist for diagnosing why a proposed multimodal fusion fails: a mismatch between the chosen principle and the actual relationship in the data would predict poor fusion.
  • The framework predicts that the same data transformation algorithm is not universally applicable; data of the same modality may need different transformations depending on the application and the alignment links.
  • The distinction between precise and coarse fusion paradigms gives a decision rule: choose precise fusion when the problem is well understood and data is scarce, and coarse fusion when the problem is vague and data is abundant.
  • The framework implies that cross-domain fusion can reduce data collection costs and improve forecast accuracy, anomaly detection earliness, and estimation reliability, as demonstrated in the urban-computing examples it surveys.

Reading between the lines

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

  • One extension the paper leaves implicit is that the four alignment principles could be formalized as a testable typology: given two datasets, a practitioner could check which principle applies by measuring the statistical relationship (shared objects, shared categories, dependencies, or shared latent structure) before committing to a fusion architecture.
  • A practical consequence not spelled out in the paper is that the framework suggests a natural diagnostic for negative transfer: if the presumed link between causal factors is wrong, fusing the data should degrade performance relative to a single-domain baseline, and the framework currently offers no method to detect such situations in advance.
  • The paper's claim that existing deep encoders fail to preserve spatial and temporal properties for spatio-temporal data points toward a concrete research program: designing representation learning algorithms that explicitly maintain distance, hierarchy, and periodicity in latent space, which would directly affect the Data Layer of this framework.
  • The framework could be extended from a descriptive methodology to a prescriptive one by quantifying the strength of each link (e.g., correlation or mutual information between datasets) and setting a threshold beyond which fusion is beneficial; this is an inference from the paper's emphasis on the strength of links.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 6 minor

Summary. The paper defines cross-domain multimodal data fusion as a research theme distinct from single-domain fusion, arguing that when data in a target domain is insufficient, knowledge from datasets generated in other domains should be fused. It proposes a four-layer framework (Domains, Links, Models, Data) with ten procedures, four knowledge alignment principles (multiview-based, similarity-based, dependency-based, commonality-based), two fusion paradigms (precise and coarse), and a data-transformation architecture. The stated contribution is a methodology answering three questions: what to fuse, why fusion is possible, and how to fuse. The paper is explicitly positioned as neither a new algorithm nor a conventional survey, but a systematization of a research direction primarily illustrated through the author's prior work on urban computing and spatio-temporal data mining.

Significance. If the framework were made operational, it could serve as a useful conceptual organization for an emerging area that sits between multimodal machine learning and application-driven physical-world problem solving. The paper's strengths are its clear problem statement, detailed running examples, and explicit acknowledgment of open questions in Section 8. It does not provide experimental validation, formal theorems, or reusable code, so its value rests entirely on the cogency of its taxonomy and the utility of its distinctions. The framework's classificatory power is real but limited by the absence of criteria for applying its principles and by the unsupported 'effectively' in the central claim linking link discovery to fusion performance.

major comments (5)
  1. [§4.1, procedures ⑤–⑥ and §8] The central claim that the framework lets us 'design solutions that fuse cross-domain multimodal data effectively' depends on the accuracy of the links identified in procedures ⑤ and ⑥. The paper states in procedure ⑤ that only 'coarse interactions between causal factors based on domain knowledge' can be obtained, and Section 8 lists 'evaluating the strength of identified links between data' and 'under what conditions does a cross-domain multimodal data fusion paradigm outperform single-domain models' as open problems. Thus the one component that distinguishes this framework from single-domain fusion is explicitly left without a correctness criterion, and there is no mechanism for detecting or preventing negative transfer when links are spurious. This is load-bearing: without a link-validation step, the framework's effectiveness claim is an assertion rather than a consequence of the proposed methodology.
  2. [§5.1.1 and §5.4.1] The set-theoretic characterizations of knowledge fusion, namely L_XY = L_X ∪ L_Y for the multiview-based principle and L_XY = L_X ∩ L_Y for the commonality-based principle, are informal and not defined at the level of actual representations. The symbols L_X, L_Y, 'overlap', and 'disjoint' are used heuristically, and the claims that disjoint views maximize fused knowledge and that overlap implies redundancy are asserted without formal justification or empirical demonstration. Since these equations are the paper's principal explanatory device for why fusion works, they need either a precise mathematical interpretation or an explicit caveat that they are only intuitive illustrations.
  3. [§5.1.2, §5.2.2, §5.3.2, §5.4.2, and §5.5] There is a circularity in the way the four alignment principles are validated: the principles are induced from the same model families (co-training, coupled matrix factorization, CRF/attention, multitask/transfer/meta-learning) that are then cited as implementations of those principles. As a result, every model in the cited families is assigned to a principle by construction, and the paper gives no decision procedure for determining which principle a new model instantiates, nor any criterion that could falsify the taxonomy. The differences between principles are described in Section 5.5, but they are not operational; a reader cannot tell, for a previously unseen fusion method, whether it is multiview-based or dependency-based without relying on the author's label.
  4. [§3.1] The claimed benefits of cross-domain knowledge fusion—more accurate forecasts, earlier anomaly detection, and more reliable estimation—are supported by citations to specific prior systems (e.g., [56][70], [72], [42][50]) rather than by evidence or analysis in this article. For a position paper, using prior systems as motivation is acceptable, but the text presents these as benefits of cross-domain knowledge fusion in general, and by implication of the proposed framework, without noting that none of the cited systems uses the four-layer framework as specified here. The distinction should be made explicit so that the framework's own effectiveness is not conflated with the success of earlier methods.
  5. [Abstract and §8] The abstract's assertion that the framework enables the design of solutions that 'fuse cross-domain multimodal data effectively' overstates what the paper establishes. Section 8 explicitly identifies as future work the evaluation of link strength and the conditions under which cross-domain fusion outperforms single-domain models. Given that the paper is a position piece, the lack of experiments is not itself fatal, but the wording of the central claim should be qualified to reflect that the framework provides a systematic procedure without yet demonstrating that the procedure improves fusion outcomes.
minor comments (6)
  1. [§1] The word 'Alternative' at the start of the paragraph describing Figure 1 B) should be 'Alternatively'.
  2. [§2.1] 'structed data' should be 'structured data' in the sentence 'Typical examp les of structed data are digital spreadsheets'.
  3. [§6.3] 'Not that what Figure 15 presents is just an example' should be 'Note that what Figure 15 presents is just an example'.
  4. [§3.2] 'very spare' should be 'very sparse' in the description of matrices in the similarity-based principle.
  5. [§5.4.2] The abbreviation 'MLT' is used for multitask learning, while the surrounding text and related references use 'MTL'; the abbreviation should be consistent.
  6. [Table 1] The table uses checkmarks and blank cells without a legend explaining that a checkmark denotes applicability; also, the column alignment is hard to read in the current text and should be reformatted.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the paper is an organizing framework with independent prior-work citations; its open questions are acknowledged gaps, not circular steps.

full rationale

This is a position/framework paper rather than a derivation of quantitative predictions. The four-layer framework (Domains, Links, Models, Data) is offered as a methodology for selecting data, identifying alignment principles, choosing fusion paradigms, and transforming representations; procedures 1-10 are procedural heuristics, not equations, and no component of the framework is defined in terms of the outcome it is claimed to explain. The four knowledge-alignment principles are abstract categories (multiview, similarity, dependency, commonality) illustrated with existing model families; classifying those families under the principles is an organizational contribution, not a circular derivation. Benefit claims in Section 3.1 and examples in Section 5 cite the author's earlier empirical papers (e.g., [70], [42], [56], [72]), but those are independently published, externally falsifiable results that predate this framework; the citations are self-citations but not load-bearing in the sense of importing an unverified premise from the present paper. The paper explicitly lists as future work 'evaluating the strength of identified links between data' and 'under what conditions does a cross-domain multimodal data fusion paradigm outperform single-domain models' (Section 8), which is an acknowledged limitation and an evidential gap, not a circular step. No fitted parameter is renamed as a prediction, no uniqueness theorem is imported from the author's prior work, and no ansatz is smuggled in via citation. Therefore no circular step can be exhibited, and the appropriate score is 0.

Assumptions & free parameters 0 free parameters · 4 assumptions · 6 invented entities

The framework rests on domain knowledge and on the assumption that the four alignment principles can be reliably instantiated. No free parameters are fitted because the paper contains no experiments or numerical claims. The principles and paradigms are conceptual inventions without independent falsifiable evidence.

assumptions (4)
  • domain assumption Domain knowledge about a problem's root causes and contributing factors is available and can be reliably elicited.
    Procedures 1 to 4 in Section 4.1 depend on this to select relevant data; the air-quality example in Section 4.2 illustrates but does not validate it.
  • domain assumption Meaningful links between disparate datasets can be identified through multiview, similarity, dependency, or commonality principles, and these links improve task performance.
    Section 5 asserts this philosophy; the paper provides no formal condition or negative-result analysis.
  • ad hoc to paper The set-theoretic operations on latent representations (such as L_XY = L_X union L_Y and L_XY = L_X intersect L_Y) are a valid model of knowledge fusion.
    Used informally in Sections 5.1.1 and 5.4.1 to justify the principles; no definitions of L_X or overlap are given.
  • ad hoc to paper The four alignment principles are sufficient to span the space of cross-domain fusion methods.
    Section 5.5 discusses differences and combinations but the paper itself concedes in Section 8 that more principles may exist.
invented entities (6)
  • Multiview-based knowledge alignment principle
    purpose: Explains fusion of multiple views of the same object across domains
    Introduced in Section 5.1; no falsifiable prediction, re-describes existing multiview learning.
  • Similarity-based knowledge alignment principle
    purpose: Explains fusion through similarity between objects of the same category
    Introduced in Section 5.2; mirrors collaborative filtering and contrastive learning, no independent evidence.
  • Dependency-based knowledge alignment principle
    purpose: Explains fusion through probabilistic dependencies between properties
    Introduced in Section 5.3; mirrors graphical models and attention, no independent evidence.
  • Commonality-based knowledge alignment principle
    purpose: Explains fusion through shared commonality between domains
    Introduced in Section 5.4; mirrors transfer and meta-learning, no independent evidence.
  • Precise and coarse knowledge fusion paradigms
    purpose: Classifies fusion algorithms by knowledge granularity and connection explicitness
    Introduced in Section 6; conceptual taxonomy without empirical validation.
  • Four-layer framework (Domains, Links, Models, Data)
    purpose: Guides design of cross-domain fusion solutions
    Introduced in Section 4; no external validation, framework's utility is asserted.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Fusing Cross-Domain Knowledge from Multimodal Data to Solve Problems in the Physical World." pith.science (2026). https://pith.science/paper/CIE24RMZ

@misc{pith2026250603155,
  author       = {Pith},
  title        = {Pith review of: Fusing Cross-Domain Knowledge from Multimodal Data to Solve Problems in the Physical World},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CIE24RMZ}},
  note         = {Machine review of arXiv:2506.03155}
}
read the original abstract

The proliferation of artificial intelligence has enabled a diversity of applications that bridge the gap between digital and physical worlds. As physical environments are too complex to model through a single information acquisition approach, it is crucial to fuse multimodal data generated by different sources, such as sensors, devices, systems, and people, to solve a problem in the real world. Unfortunately, it is neither applicable nor sustainable to deploy new resources to collect original data from scratch for every problem. Thus, when data is inadequate in the domain of problem, it is vital to fuse knowledge from multimodal data that is already available in other domains. We call this cross-domain knowledge fusion. Existing research focus on fusing multimodal data in a single domain, supposing the knowledge from different datasets is intrinsically aligned; however, this assumption may not hold in the scenarios of cross-domain knowledge fusion. In this paper, we formally define the cross-domain multimodal data fusion problem, discussing its unique challenges, differences and advantages beyond data fusion in a single domain. We propose a four-layer framework, consisting of Domains, Links, Models and Data layers, answering three key questions:"what to fuse", "why can be fused", and "how to fuse". The Domains Layer selects relevant data from different domains for a given problem. The Links Layer reveals the philosophy of knowledge alignment beyond specific model structures. The Models Layer provides two knowledge fusion paradigms based on the fundamental mechanisms for processing data. The Data Layer turns data of different structures, resolutions, scales and distributions into a consistent representation that can be fed into an AI model. With this framework, we can design solutions that fuse cross-domain multimodal data effectively for solving real-world problems.

Figures

Figures reproduced from arXiv: 2506.03155 by the authors.

Figure 2
Figure 2. Illustration of cross-domain knowledge fusion 2) Problems in Digital World vs in Physical World: There are three scenarios of problem solving in which we fuse multimodal data, as illustrated in [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

76 extracted references · 64 canonical work pages

  1. [56]

    Xiuwen Yi, Junbo Zhang, Zhaoyuan Wang, Tianrui Li, and Yu Zheng. 2018. Deep distributed fusion network for air quality prediction. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD’18), 965-973

  2. [70]

    Yu Zheng, Xiuwen Yi, Ming Li, Yanhua Li, Zhangqing Shan, Eric Y Chang, and Tianrui Li

  3. [72]

    Yu Zheng, Huichu Zhang, and Yong Yu. 2015. Detecting collective anomalies from multiple 1: 34 ● Y. Zheng ACM Trans. Intelligent systems and technologies, Vol. x, No. x, Article 1, Pub. date: July 2025. spatio-temporal datasets across different domains. In Proceedings of the 23rd SIGSPATIAL International Conference on Advances in Geographic Information Sys...

  4. [42]

    Jingbo Shang, Yu Zheng, Wenzhu Tong, Eric Chang, and Yong Yu. 2014. Inferring gas consumption and pollution emission of vehicles throughout a city. In Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD’14), 1027-1036

  5. [50]

    Yilun Wang, Yu Zheng, and Yexiang Xue. 2014. Travel time estimation of a path using sparse trajectories. In Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD’14), 25-34

  6. [1]

    Abdu, Ahmed H

    Sarah A. Abdu, Ahmed H. Yousef, and Ashraf Salem. 2021. Multimodal video sentiment analysis using deep learning approaches, a survey. Information Fusion 76, 12 (2021), 204-226

  7. [2]

    Felipe Almeida and Geraldo Xexé o. 2023. Word embeddings: A survey. arXiv:1901.09069. Retrieved from https://arxiv.org/abs/1901.09069

  8. [3]

    Dor Bank, Noam Koenigstein, and Raja Giryes. 2021. Autoencoders. arXiv:2003.05991. Retrieved from https://arxiv.org/abs/2003.05991

Show all 76 references
  1. [4]

    Jonathan Baxter. 2000. A model of inductive bias learning. Journal of Artificial Intelligence Research 12, 1 (2000), 149-198

  2. [5]

    Yoshua Bengio, Ré jean Ducharme, Pascal Vincent, and Christian Janvin. 2003. A neural probabilistic language model. The Journal of Machine Learning Research 3, (2003), 1137-1155

  3. [6]

    Blei, Andrew Y

    David M. Blei, Andrew Y. Ng, and Michael I. Jordan. 2003. Latent dirichlet allocation. The Journal of Machine Learning Research 3, (2003), 993-1022

  4. [7]

    Avrim Blum and Tom Mitchell. 1998. Combining labeled and unlabeled data with co -training. In Proceedings of the 11th Annual Conference on Computational Learning Theory (July 1998)

  5. [8]

    Tim Brooks, Bill Peebles, Connor Holmes, Will DePue, Yufei Guo, Li Jing, David Schnurr, Joe Taylor, Troy Luhman, Eric Luhman, Clarence Ng, Ricky Wang, and Aditya Ramesh. 2024. Video generation models as world simulators. https://openai.com/research/video-generation- models-as-...

  6. [9]

    Rich Caruana. 1993. Multitask learning: A knowledge -based source of inductive bias. In Proceedings of the 10th International Conference on International Conference on Machine Learning (ICML’93), 41-48

  7. [10]

    Ning Chen, Jun Zhu, and Eric P. Xing. 2010. Predictive subspace learning for multi-view data: A large margin approach. In Proceedings of the 24th International Conference on Neural Information Processing Systems (NIPS’10), 361-369

  8. [11]

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. 2020. A simple framework for contrastive learning of visual representations. In Proceedings of the 37th International Conference on Machine Learning (ICML’20), Article 149, 1597-1607

  9. [12]

    Peng Cui, Xiao Wang, Jian Pei, and Wenwu Zhu. 2019. A survey on network embedding. IEEE Transactions on Knowledge and Data Engineering 31, 5 (2019), 833-852

  10. [13]

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human...

  11. [14]

    Darren Edge, Ha Trinh, Newman Cheng, Joshua Bradley, Alex Chao, Apurva Mody, Steven Truitt, and Jonathan Larson. 2024. From local to global: A graph RAG approach to query - focused summarization. arXiv:2404.16130. Retrieved from https://arxiv.org/abs/2404.16130

  12. [15]

    Jing Gao, Peng Li, Zhikui Chen, and Jianing Zhang. 2020. A survey on deep learning for multimodal data fusion. Neural Computation 32, 5 (2020), 829-864

  13. [16]

    Hassan Gharoun, Fereshteh Momenifar, Fang Chen, and Amir H. Gandomi. 2024. Meta - learning approaches for few -shot learning: A survey of recent advances. ACM Computing Surveys 56, 12, Article 294 (2024), 1-41

  14. [17]

    Mehmet Gö nen and Ethem Alpaydin. 2011. Multiple kernel learning algorithms. Journal of Machine Learning Research 12, (2011), 2211-2268

  15. [18]

    Desta Haileselassie Hagos, Rick Battle, and Danda B. Rawat. 2024. Recent advances in generative AI and large language models: Current status, challenges, and perspectives. IEEE Transactions on Artificial Intelligence 5, 12 (2024), 5873-5893. Fusing Cross-Domain Knowledge from ...

  16. [19]

    Hoang, Yu Zheng, and Ambuj K

    Minh X. Hoang, Yu Zheng, and Ambuj K. Singh. 2016. FCCF: Forecasting citywide crowd flows based on big data. In Proceedings of the 24th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems (SIGSPATIAL’16). Article 6, 1 - 10

  17. [20]

    Timothy Hospedales, Antreas Antoniou, Paul Micaelli, and Amos Storkey. 2022. Meta-learning in neural networks: A survey. IEEE Transactions on Pattern Analysis & Machine Intelligence 44, 9 (2022), 5149-5169

  18. [21]

    Guangyin Jin, Yuxuan Liang, Yuchen Fang, Zezhi Shao, Jincai Huang, Junbo Zhang, and Yu Zheng. 2024. Spatio-temporal graph neural networks for predictive learning in urban computing: A survey. IEEE Transactions on Knowledge and Data Engineering 36, 10 (2024), 5388-5408

  19. [22]

    Dana Lahat, Tü lay Adali, and Christian Jutten. 2015. Multimodal data fusion: An overview of methods, challenges, and prospects. In Proceedings of the IEEE 103, 9 (2015), 1449-1477

  20. [23]

    Christiane Lemke, Marcin Budka, and Bogdan Gabrys. 2013. Metalearning: A survey of trends and technologies. Artificial Intelligence Review 44, 1 (2013), 117-130

  21. [24]

    Rongfan Li, Ting Zhong, Xinke Jiang, Goce Trajcevski, Jin Wu, and Fan Zhou. 2022. Mining spatio-temporal relations via self-paced graph contrastive learning. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD’22), 936-944

  22. [25]

    Yuxuan Liang, Songyu Ke, Junbo Zhang, Xiuwen Yi, and Yu Zheng. 2018. GeoMAN: Multi - level attention networks for geo -sensory time series prediction. In Proceedings of the 27th International Joint Conference on Artificial Intelligence (IJCAI’18), 3428-3434

  23. [26]

    Yang Liu, Weixing Chen, Yongjie Bai, Jingzhou Luo, Xinshuai Song, Kaixuan Jiang, Zhida Li, Ganlong Zhao, Junyi Lin, Guanbin Li, Wen Gao, and Liang Lin. 2024. Aligning cyber space with physical world: A comprehensive survey on embodied AI. arXiv:2407.06886. Retrieved from https...

  24. [27]

    Rosenblum

    Ye Liu, Yu Zheng, Yuxuan Liang, Shuming Liu, and David S. Rosenblum. 2016. Urban water quality prediction based on multi -task multi -view learning. In Proceedings of the 25th International Joint Conference on Artificial Intelligence (IJCAI’16), 2576-2582

  25. [28]

    Xu Liu, Yuxuan Liang, Chao Huang, Yu Zheng, Bryan Hooi, and Roger Zimmermann. 2022. When do contrastive learning signals help spatio-temporal graph forecasting? In Proceedings of the 30th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems (SI...

  26. [29]

    Yin Lou, Chengyang Zhang, Yu Zheng, Xing Xie, Wei Wang, and Yan Huang. 2009. Map - matching for low-sampling-rate GPS trajectories. In Proceedings of the 17th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems (GIS’09), 352-361

  27. [30]

    Yueen Ma, Zixing Song, Yuzheng Zhuang, Jianye Hao, and Irwin King. 2024. A survey on vision-language-action models for embodied AI. arXiv:2405.14093. Retrieved from https://arxiv.org/abs/2405.14093

  28. [31]

    Jonathan Masci, Ueli Meier, Dan Cireşan, and Jürgen Schmidhuber. 2011. Stacked convolutional auto -encoders for hierarchical feature extraction. In Proceedings of the 21th International Conference on Artificial Neural Networks - Volume Part I (ICANN’11), 52-59

  29. [32]

    Tong Meng, Xuyang Jing, Zheng Yan, and Witold Pedrycz. 2020. A survey on machine learning for data fusion. Information Fusion 57, 5 (2020), 115-129

  30. [33]

    Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean. 2013. Efficient estimation of word representations in vector space. Retrieved from https://arxiv.org/abs/1301.3781

  31. [34]

    Siyuan Mu and Sen Lin. 2025. A Comprehensive Survey of Mixture-of-Experts: Algorithms, Theory, and Applications. arXiv:2503.07137

  32. [35]

    Sinno Jialin Pan and Qiang Yang. 2010. A survey on transfer learning. IEEE Transactions on Knowledge and Data Engineering 22, 10 (2010), 1345-1359

  33. [36]

    Zheyi Pan, Wentao Zhang, Yuxuan Liang, Weinan Zhang, Yong Yu, Junbo Zhang, and Yu Zheng. 2022. Spatio-temporal meta learning for urban traffic prediction. IEEE Transactions on Knowledge and Data Engineering 34, 3 (2022), 1462-1476. 1: 32 ● Y. Zheng ACM Trans. Intelligent syste...

  34. [37]

    Huiling Qin, Songyu Ke, Xiaodu Yang, Haoran Xu, Xianyuan Zhan, and Yu Zheng. 2021. Robust spatio-temporal purchase prediction via deep meta learning. In Proceedings of the AAAI Conference on Artificial Intelligence 35, 5 (2021), 4312-4319

  35. [38]

    Hao Qu, Yongshun Gong, Meng Chen, Junbo Zhang, Yu Zheng, and Yilong Yin. 2023. Forecasting fine -grained urban flows via spatio -temporal contrastive self -supervision. IEEE Transactions on Knowledge and Data Engineering 35, 8 (2023), 8008-8023

  36. [39]

    Aditya Ramesh, Mikhail Pavlov, Gabriel Goh, Scott Gray, Chelsea Voss, Alec Radford, Mark Chen, and Ilya Sutskever. 2021. Zero-shot text-to-image generation. In Proceedings of the 38th International Conference on Machine Learning (ICML’21), 8821-8831

  37. [40]

    Sö ren Richard Stahlschmidt, Benjamin Ulfenborg, and Jane Synnergren. 2022. Multimodal deep learning for biomedical data fusion: A review. Briefings in Bioinformatics 23, 2 (2022), 1- 15

  38. [41]

    Sijie Ruan, Cheng Long, Zhipeng Ma, Jie Bao, Tianfu He, Ruiyuan Li, Yiheng Chen, Shengnan Wu, and Yu Zheng. 2022. Service time prediction for delivery tasks via spatial meta -learning. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD...

  39. [43]

    Singh and Geoffrey J

    Ajit P. Singh and Geoffrey J. Gordon. 2008. Relational learning via collective matrix factorization. In Proceedings of the 14th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD’08), 650-658

  40. [44]

    Amanpreet Singh and Ronghang Hu and Vedanuj Goswami and Guillaume Couairon and Wojciech Galuba and Marcus Rohrbach and Douwe Kiela . 2022. FLAVA: A Foundational Language And Vision Alignment Model. arXiv:2112.04482

  41. [45]

    Junkai Sun, Junbo Zhang, Qiaofei Li, Xiuwen Yi, Yuxuan Liang, and Yu Zheng. 2022. Predicting citywide crowd flows in irregular regions using multi -view graph convolutional networks. IEEE Transactions on Knowledge and Data Engineering 34, 5 (2022), 2348-2359

  42. [46]

    Yu, and Tianyi Wu

    Yizhou Sun, Jiawei Han, Xifeng Yan, Philip S. Yu, and Tianyi Wu. 2022. Heterogeneous information networks: The past, the present, and the future. In Proceedings of the VLDB Endowment 15, 12 (2022), 3807-3811

  43. [47]

    Masahiro Suzuki and Yutaka Matsuo. 2022. A survey of multimodal deep generative models. Advanced Robotics 36, 5-6 (2022), 261-278

  44. [48]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. Attention Is All You Need. arXiv:1706.03762. Retrieved from https://arxiv.org/abs/1706.03762

  45. [49]

    Senzhang Wang, Jiannong Cao, and Philip S. Yu. 2022. Deep learning for spatio-temporal data mining: A survey. IEEE Transactions on Knowledge and Data Engineering 34, 8 (2022), 3681- 3700

  46. [51]

    Chang Xu, Dacheng Tao, and Chao Xu. 2013. A survey on multi -view learning. arXiv:1304.5634. Retrieved from https://arxiv.org/abs/1304.5634

  47. [52]

    Zhiyuan Xu, Kun Wu, Junjie Wen, Jinming Li, Ning Liu, Zhengping Che, and Jian Tang. 2024. A survey on robotics with foundation models: Toward embodied AI. arXiv:2402.02385. Retrieved from https://arxiv.org/abs/2402.02385

  48. [53]

    Xiaoqiang Yan, Shizhe Hu, Yiqiao Mao, Yangdong Ye, and Hui Yu. 2021. Deep multi -view learning methods: A review. Neurocomputing 448, 30 (2021), 106–129

  49. [54]

    Qiang Yang, Yuqiang Chen, Gui-Rong Xue, Wenyuan Dai, and Yong Yu. 2009. Heterogeneous Fusing Cross-Domain Knowledge from Multimodal Data ACM Trans. Intelligent systems and technologies, Vol. x, No. x, Article 1, Pub. date: July 2025. transfer learning for image clustering via ...

  50. [55]

    Quanming Yao, Mengshuo Wang, Yuqiang Chen, Wenyuan Dai, Yi -Qi Hu, Yu-Feng Li, Wei- Wei Tu, Qiang Yang, and Yang Yu. 2018. Taking human out of learning applications: A survey on automated machine learning. Retrieved from https://arxiv.org/abs/1810.13306v3

  51. [57]

    Yuning You, Tianlong Chen, Yongduo Sui, Ting Chen, Zhangyang Wang, and Yang Shen. 2020. Graph contrastive learning with augmentations. In Proceedings of the 34th International Conference on Neural Information Processing Systems (NIPS’20), Article 488, 5812-5823

  52. [58]

    Jing Yuan, Yu Zheng, and Xing Xie. 2012. Discovering regions of different functions in a city using human mobility and POIs. In Proceedings of the 18th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD’12), 186-194

  53. [59]

    Nicholas Jing Yuan, Yu Zheng, Xing Xie, Yingzi Wang, Kai Zheng, and Hui Xiong. 2015. Discovering urban functional zones using latent activity trajectories. IEEE Transactions on Knowledge and Data Engineering 27, 3 (2015), 712-725

  54. [60]

    Duzhen Zhang, Yahan Yu, Chenxing Li, Jiahua Dong, Dan Su, Chenhui Chu, and Dong Yu

  55. [61]

    Junbo Zhang, Yu Zheng, and Dekang Qi. 2017. Deep spatio -temporal residual networks for citywide crowd flows prediction. In Proceedings of the AAAI Conference on Artificial Intelligence 31, 1 (2017)

  56. [62]

    Junbo Zhang, Yu Zheng, Junkai Sun, and Dekang Qi. 2020. Flow prediction in spatio-temporal networks based on multitask deep learning. IEEE Transactions on Knowledge and Data Engineering 32, 3 (2020), 468-478

  57. [63]

    Yifei Zhang, Dé siré Sidibé , Olivier Morel, and Fabrice Mé riaudeau. 2021. Deep multimodal fusion for semantic image segmentation: A survey. Image and Vision Computing 105, 1 (2021)

  58. [64]

    Fei Zhao, Chengcui Zhang, and Baocheng Geng. 2024. Deep multimodal data fusion. ACM Computing Surveys 56, 9, Article 216 (2024), 1-36

  59. [65]

    Yu Zheng. 2015. Methodologies for cross-domain data fusion: An overview. IEEE Transactions on Big Data 1, 1 (2015), 16-34

  60. [66]

    Yu Zheng. 2019. Urban Computing. MIT Press, Cambridge

  61. [67]

    Yu Zheng, Licia Capra, Ouri Wolfson, and Hai Yang. 2014. Urban computing: Concepts, methodologies and applications. ACM Transactions on Intelligent Systems and Technology 5, 3 (2014), 1-55

  62. [68]

    Yu Zheng, Furui Liu, and Hsun -Ping Hsieh. 2013. U -Air: When urban air quality inference meets big data. In Proceedings of the 19th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD’13), 1436-1444

  63. [69]

    Yu Zheng, Tong Liu, Yilun Wang, Yanmin Zhu, Yanchi Liu, and Eric Chang. 2014. Diagnosing New York city’s noises with ubiquitous data. In Proceedings of the 2014 ACM International Joint Conference on Pervasive and Ubiquitous Computing (UbiComp’14), 715-725

  64. [71]

    Yu Zheng, Xiuwen Yi, Dekang Qi, and Zheyi Pan. 2024. Elementarisation method for public data based on urban knowledge systems. Big Data Research 10, 4 (2024), 130-148

  65. [73]

    Yu Zheng, Lizhu Zhang, Xing Xie, and Wei -Ying Ma. 2009. Mining interesting locations and travel sequences from GPS trajectories. In Proceedings of the 18th International Conference on World Wide Web (WWW’09), 791-800

  66. [74]

    Zheng, Yu Zheng, Xing Xie, and Qiang Yang

    Vincent W. Zheng, Yu Zheng, Xing Xie, and Qiang Yang. 2010. Collaborative location and activity recommendations with GPS history data. In Proceedings of the 19th International Conference on World Wide Web (WWW’10), 1029-1038. Dr. Yu Zheng is the Vice President of JD.COM and pr...

  67. [2015]

    In Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD’15) , 2267-2276

    Forecasting fine-grained air quality based on big data. In Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD’15) , 2267-2276

  68. [2024]

    In Findings of the Association for Computational Linguistics: ACL 2024

    MM-LLMs: Recent advances in multimodal large language models. In Findings of the Association for Computational Linguistics: ACL 2024 . Association for Computational Linguistics, Bangkok, Thailand, 12401-12430

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.