Pith. sign in

REVIEW 2 major objections 6 minor 40 references

MobiCLR: Mobility Time Series Contrastive Learning for Urban Region Representations

T0 review · 2 major / 6 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read The paper claims that contrastive learning on hourly inbound and outbound trip counts yields urban region embeddings that linearly predict income, education, and social vulnerability across three US cities.

desk verdict A plausible, modestly novel mobility-contrastive architecture whose headline 'wins in all cells' claim outruns the statistics and whose evaluation protocol is not clearly separated from the test set; fixable and worth engaging. read the letter →

arxiv 2502.02912 v1 pith:UWN7EUJ2 submitted 2025-02-05 cs.LG cs.AI

Namwoo Kim , Takahiro Yabe , Chanyoung Park , Yoonjin Yoon This is my paper · ORCID
classification cs.LGcs.AI
keywords urbanregionembeddingtimeseriescontrastivelearninghumanmobilityinflowoutflowpredictionsocialvulnerabilityindextransfertaxitripdata
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that urban region embeddings learned from hourly inbound and outbound trip counts via contrastive time-series representation learning can linearly predict income, educational attainment, and social vulnerability better than prior unsupervised and non-neural baselines. It argues that treating inbound and outbound flows with separate instance-wise contrastive losses, then aligning the combined representation with both flow views, captures temporal dynamics that static origin-destination or single-stream models miss. The claimed payoff is that two weeks of taxi and ride-hailing data suffice to produce transferable region embeddings for socioeconomic prediction in Chicago, New York, and Washington D.C.

What carries the argument

The central object is the three-encoder contrastive objective with two instance-wise NT-Xent losses, one per flow direction, plus an auxiliary regularizer that pulls the pooled combined embedding toward the pooled inbound and outbound embeddings. The separate losses let each flow direction develop its own temporal semantics; the regularizer forces the unified embedding to retain both views, and the design choice to discard the flow encoders at test time means the downstream representation is the one aligned with both flow-specific views.

What would settle it

Retrain MobiCLR on the same cities using either a full year of trip records or all transportation modes rather than taxis alone, and check whether it still beats TS2Vec and the other baselines on social vulnerability prediction; if the advantage shrinks or reverses, the two-week taxi-window premise is the real driver of the reported results.

Watch

Extended reading notes

Core claim

MobiCLR learns a region representation by training three dilated-convolution encoders: one on inbound trips, one on outbound trips, and one on the concatenated inbound/outbound series. The two flow-specific encoders are trained with NT-Xent contrastive losses comparing augmented views at each time step; the combined encoder is trained with an auxiliary regularizer that aligns its temporally pooled embedding with each flow-specific pooled embedding. After pre-training, only the combined encoder is kept, its representations are average-pooled, and a ridge regressor predicts the target indicator. The paper reports that this outperforms raw inflow/outflow counts, a graph attention network on static OD flows, Mixing-up, TS-TCC, and TS2Vec on all nine city-task combinations, with the largest gains on social vulnerability.

Load-bearing premise

The whole comparison rests on the assumption that two weeks of taxi and ride-hailing trips in each city capture the stable mobility patterns of residents closely enough that the learned embeddings predict income, education, and vulnerability; the paper itself notes that only about 17% of commuters use these modes.

Editorial extensions

If this is right

  • If the central claim is right, two weeks of taxi and ride-hailing trip records are enough to estimate neighborhood-level socioeconomic indicators without relying on census questions.
  • The transferability tests suggest a model trained on one city's mobility data can predict social vulnerability in another city, so the approach may work where such indicators have not yet been measured.
  • The performance gap over static OD graph models supports the paper's assertion that temporal dynamics carry signal that aggregated flow counts lose.
  • The ablation study indicates the auxiliary regularizer is load-bearing: removing it drops the average social vulnerability R-squared from 0.616 to 0.565.
  • The same pretraining protocol extends to composite indices like social vulnerability, not just single indicators such as income or education.

Reading between the lines

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

  • The paper's own limitation suggests a direct extension: applying MobiCLR to multimodal transport data (bus, subway, and walking) could either confirm robustness or reveal that taxi-only signals systematically skew toward wealthier neighborhoods.
  • Because the regularizer only aligns pooled summary vectors, a natural test would be whether a single two-channel encoder with the same total loss matches MobiCLR; if it does, the architectural separation into three encoders is not the essential ingredient.
  • The reported R-squared values vary widely by city, so the claim of universal superiority may be sensitive to each city's taxi mode share and spatial granularity, a pattern a broader multi-city study could map out.
  • The two-week window is a single spring period; repeating the experiment on other seasons would test whether the learned temporal semantics are stable across the year or specific to that April window.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 6 minor

Summary. The paper proposes MobiCLR, an unsupervised method for learning urban region embeddings from hourly inbound and outbound mobility time series. The model uses three encoders to obtain inbound-specific, outbound-specific, and combined representations, trains them with instance-wise contrastive losses, and adds an auxiliary regularizer that aligns the combined representation with the two flow-specific views. The authors evaluate the resulting embeddings by linear probing on American Community Survey income and educational attainment and on CDC Social Vulnerability Index in Chicago, New York, and Washington, D.C., comparing against GAT, raw flow features, Mixing-up, TS-TCC, and TS2Vec. They report R^2 values averaged over five runs, provide ablation, sensitivity, and transferability analyses, and conclude that MobiCLR outperforms the tested baselines across all downstream tasks in all three cities.

Significance. The core idea is reasonable and potentially useful: capturing temporal dynamics of inbound and outbound flows separately and then aligning them into a unified representation is a natural extension of prior contrastive region-embedding work. The evaluation uses external, semantically meaningful benchmarks (ACS and CDC) rather than circular internal reconstruction tasks, and the paper includes ablations, transfer tests, and reproducible open-data sources. If the reported gains survive a properly controlled evaluation, the method would be a solid contribution to urban region representation learning. However, the current evidence is weakened by the evaluation protocol: augmentation and regularization choices appear to be selected using test-set outcomes, and the headline claim of superiority across all tasks is not supported by significance testing, with several differences within one standard deviation.

major comments (2)
  1. [§6.1 and §5.4] The comparison is not yet head-to-head because there is no stated validation split. Section 6.1 selects the jitter-then-shift augmentation from Figure 3d, which reports R^2 of social vulnerability prediction, while Section 5.4 states that ridge regression is evaluated on the test split. Section 5.4 also selects the ridge regularization parameter by grid search without saying that this is a separate validation set. If the test labels influenced augmentation and regularization choices, MobiCLR's hyperparameters are effectively tuned on the test set, whereas the baselines are run with their original configurations per the Appendix. Please introduce an explicit validation split or nested cross-validation for all hyperparameter and augmentation choices, and re-run the baselines under the same selection protocol, or restrict the claims accordingly.
  2. [§5.4 / Table 2] The claim that MobiCLR 'outperformed non-neural network models and state-of-the-art unsupervised methods across all downstream tasks in the three cities' is stronger than Table 2 supports. Several gaps are within one standard deviation of the runner-up: NY EDU 0.790 ± 0.015 vs TS2Vec 0.789 ± 0.017; CH EDU 0.759 ± 0.020 vs TS2Vec 0.746 ± 0.020; NY INC 0.817 ± 0.015 vs TS2Vec 0.804 ± 0.012; and NY SVI 0.792 ± 0.009 vs TS2Vec 0.785 ± 0.017. No paired significance test or confidence interval is reported. In addition, the stated SVI improvements over runner-up models (12.54% for Chicago, 31.56% for Washington, D.C., and 3.13% for New York) do not match Table 2: for DC SVI the runner-up is Mixing-up at 0.363, giving 22.9%, and for NY SVI the runner-up is TS2Vec at 0.785, giving 0.9%. Please report paired significance tests across the five runs and recalibrate the claims to the cells where the model is actually better.
minor comments (6)
  1. [Appendix (Baselines)] The baseline descriptions for the raw time-series features are swapped: the 'Inbound time series (x_i)' entry says it represents raw hourly counts of outbound trips, and the 'Outbound time series (x_o)' entry says it represents inbound trips.
  2. [§5.4] The sentence 'the pooling of h^io_i across all timestamps' should be 'the pooling of h^io_n' or 'h^io_{n,*}'; the subscript i appears to be a typo.
  3. [§4.3.2] The phrase 'In contrast to Eqs. (2) and (3)' is imprecise because Eq. (3) is the inbound minibatch loss; the intended reference is likely Eqs. (1)–(4).
  4. [§6.3] The sentence 'Figure 4 presents the performance of the proposed model corresponding to different values of' is missing the parameter name; state which hyperparameter is varied in each panel.
  5. [§6.4] The statement that the model 'outperforming the second best approach in the non-transfer setting' is ambiguous; specify which baseline, which transfer direction, and which comparison cell.
  6. [Figure 3 caption] For compositions of two transformations, order may matter (jitter then shift versus shift then jitter); the caption should state how the off-diagonal entries are ordered so the matrices can be interpreted.

Circularity Check

2 steps flagged · score 6.0 of 10

Evaluation selection, not model derivation, creates partial circularity: augmentation and ridge regularization are chosen using the same test SVI outcomes later reported as predictions.

  1. fitted input called prediction [Section 6.1 (Data Augmentation), Figure 3d; combined with the test-split protocol of Section 5.1]
    "Nonetheless, a data augmentation strategy that includes shift after the jitter consistently produced robust results across all cities, as evidenced by the average R2 of social vulnerability prediction for the three cities presented in Figure 3d."

    Section 5.1 defines only a 75/25 train/test split and reports results on the test split. Section 6.1 then uses Figure 3d, the average R2 of social-vulnerability prediction across the three cities, to select the jitter+shift augmentation. The same test labels that determine the final augmentation are the labels used in Table 2's reported SVI R2 values, so the reported SVI gains (e.g., DC 0.446 vs. 0.363 runner-up, described as a 31.56% improvement) are selected outcomes rather than independent predictions.

  2. fitted input called prediction [Section 5.4 (Prediction Results), p.7; with the evaluation protocol in Section 5.1]
    "The regularization term of the linear regression model was selected through a grid search of {0.1, 0.2, 0.5, 1, 2, 5, 10}."

    Section 5.4 grid-searches the ridge regularization parameter and then reports Table 2 R2 values, but no validation split is described anywhere in Section 5.1 or 5.4. If the grid search is performed on the same test labels used to report results, the downstream linear probe is fitted to the target before 'prediction' is reported. Baselines are run with their original configurations per the Appendix, so the comparison is not a head-to-head test. The paper does not state that a separate validation split was used.

full rationale

The core model derivation is not circular: the three contrastive losses (Eqs. 1-8) are self-contained and do not reference the downstream income, education, or SVI labels, and the evaluation uses external ACS and CDC benchmarks. The only self-citation, [15], appears in the related-work survey and is not load-bearing. The circularity concern is confined to the evaluation protocol: the final augmentation is selected in Section 6.1 by inspecting average test R2 for social vulnerability, and the ridge regularization is grid-searched in Section 5.4 without any documented validation split. These choices are part of the final Table 2 configuration, so the reported 'predictions' are partly selected from the same test labels, especially for SVI; baselines are run with their original configurations, weakening the head-to-head comparison. If a hidden validation split was used, this concern would disappear, but the paper does not say so. Score 6 reflects partial evaluation-level circularity, not a derivation chain that equals its inputs by definition.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the representativeness of taxi mobility, hand-chosen augmentation and model hyperparameters, and grid-searched ridge regularization. No new physical entities are introduced.

free parameters (5)
  • Temperature parameters tau and tau_a = 1 and 0.1
    Set in Section 5.1; contrastive loss scale is sensitive to these and they are chosen by hand rather than formally tuned on a validation set.
  • Augmentation noise scales and dropout rate = 0.2 for jitter, shift, and scale; 0.1 for dropout
    Defined in Section 4.2 and the Appendix; hand-chosen magnitudes that control the invariances learned by the encoder.
  • Batch size and embedding dimension = 4 and 128
    Section 5.1; the sensitivity analysis in Figure 4 shows performance varies with these values, and the best-performing settings are used in the main results.
  • Ridge regularization parameter = selected from {0.1, 0.2, 0.5, 1, 2, 5, 10}
    Section 5.4; a grid search is described but no validation split or selection procedure is reported, so selection may use test information.
  • Augmentation composition = jitter followed by shift
    Chosen from the 4x4 augmentation grid in Figure 3 based on average SVI prediction R2 across the three cities, which is a selection on the evaluation target.
assumptions (4)
  • domain assumption Taxi and ride-hailing trip counts are a valid proxy for region-level mobility and socioeconomic characteristics.
    Used throughout Sections 5.2 and 7; the paper itself acknowledges only about 17% of commuters use taxis or ride-hailing, so the proxy can be systematically biased.
  • domain assumption Two weeks of data (336 hourly timestamps) are sufficient to represent the stable temporal mobility patterns of each region.
    Table 1 lists exactly one two-week window per city; no seasonal or multi-period validation is provided.
  • domain assumption Jittering and shifting the mobility time series preserves the label-relevant information needed for downstream income, education, and vulnerability prediction.
    Section 4.2 adopts these augmentations; the choice is validated only through downstream R2, not through any independent check of preserved semantics.
  • domain assumption The spatial join of trip pickup and drop-off points to community areas and census tracts is accurate enough for the reported R2 values.
    Section 5.2 describes the spatial join but provides no error analysis for the join or for alignment between mobility units and ACS or CDC label units.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MobiCLR: Mobility Time Series Contrastive Learning for Urban Region Representations." pith.science (2026). https://pith.science/paper/UWN7EUJ2

@misc{pith2026250202912,
  author       = {Pith},
  title        = {Pith review of: MobiCLR: Mobility Time Series Contrastive Learning for Urban Region Representations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UWN7EUJ2}},
  note         = {Machine review of arXiv:2502.02912}
}
read the original abstract

Recently, learning effective representations of urban regions has gained significant attention as a key approach to understanding urban dynamics and advancing smarter cities. Existing approaches have demonstrated the potential of leveraging mobility data to generate latent representations, providing valuable insights into the intrinsic characteristics of urban areas. However, incorporating the temporal dynamics and detailed semantics inherent in human mobility patterns remains underexplored. To address this gap, we propose a novel urban region representation learning model, Mobility Time Series Contrastive Learning for Urban Region Representations (MobiCLR), designed to capture semantically meaningful embeddings from inflow and outflow mobility patterns. MobiCLR uses contrastive learning to enhance the discriminative power of its representations, applying an instance-wise contrastive loss to capture distinct flow-specific characteristics. Additionally, we develop a regularizer to align output features with these flow-specific representations, enabling a more comprehensive understanding of mobility dynamics. To validate our model, we conduct extensive experiments in Chicago, New York, and Washington, D.C. to predict income, educational attainment, and social vulnerability. The results demonstrate that our model outperforms state-of-the-art models.

Figures

Figures reproduced from arXiv: 2502.02912 by the authors.

Figure 1
Figure 1. Overall framework of MobiCLR. We first pre-process region-wise hourly inbound and outbound trips. Subsequently, a two-step data augmentation process is applied to the mobility time series data, followed by the feeding of this augmented data into three encoders: 𝑓 𝑖 𝜃 , 𝑓 𝑜 𝜃 , and 𝑓 𝑖 𝜃 𝑜. Here, 𝑓 𝑖 𝜃 and 𝑓 𝑜 𝜃 learn inbound- and outbound-specific human mobility patterns in each region, respectively, while 𝑓 𝑖𝑜 𝜃 ex… view at source ↗
Figure 2
Figure 2. Graphical illustration of data augmentation strategy. N Kim et al.: Preprint submitted to Elsevier Page 4 of 13 [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Results of social vulnerability prediction under different time series data augmentation strategies, including individual transformations and sequential compositions of two transformations. The diagonal entries represent the results of single transformations, while the off-diagonal entries show the results of compositions of two transformations. 6.2. Ablation Study The proposed model employs a joint learning objecti… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Sensitivity analysis 6.4. Transferability Test A transferability test was conducted to assess the applicability of the learned model parameters from the source dataset to the target dataset. A model trained on mobility data from one city was utilized to predict social …
Figure 5
Figure 5. Figure 5: 𝑅2 for the transferability test on social vulnerability index. The model is trained on one source city and then evaluated on other target city. 7. Conclusion This study proposed a novel method for learning region representations using the hourly counts of inbound and o…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 36 canonical work pages

  1. [17]

    Urban region embedding via multi-view contrastive prediction, in: Proceedings of the AAAI Conference on Artificial Intelligence, pp

    Li, Z., Huang, W., Zhao, K., Yang, M., Gong, Y., Chen, M., 2024. Urban region embedding via multi-view contrastive prediction, in: Proceedings of the AAAI Conference on Artificial Intelligence, pp. 8724–8732

    work page 2024

  2. [37]

    Regionembeddingwithintraandinter-viewcontrastivelearning

    Zhang,L.,Long,C.,Cong,G.,2022. Regionembeddingwithintraandinter-viewcontrastivelearning. IEEETransactionsonKnowledgeand Data Engineering 35, 9031–9036

    work page 2022

  3. [1]

    Vulnerability

    Adger, W.N., 2006. Vulnerability. Global environmental change 16, 268–281

    work page 2006

  4. [2]

    Machinelearningandphonedatacanimprovetargetingofhumanitarian aid

    Aiken,E.,Bellue,S.,Karlan,D.,Udry,C.,Blumenstock,J.E.,2022. Machinelearningandphonedatacanimprovetargetingofhumanitarian aid. Nature 603, 864–870

    work page 2022

  5. [3]

    Mapping vulnerability: disasters, development, and people

    Bankoff, G., Frerks, G., Hilhorst, T., Hilhorst, D., 2004. Mapping vulnerability: disasters, development, and people. Routledge

    work page 2004

  6. [4]

    Community vulnerability and mobility: What matters most in spatio-temporal modeling of the covid-19 pandemic? Social Science & Medicine 287, 114395

    Carroll, R., Prentice, C.R., 2021. Community vulnerability and mobility: What matters most in spatio-temporal modeling of the covid-19 pandemic? Social Science & Medicine 287, 114395

    work page 2021

  7. [5]

    Multiple local 3d cnns for region-based prediction in smart cities

    Chen, Y., Zou, X., Li, K., Li, K., Yang, X., Chen, C., 2021. Multiple local 3d cnns for region-based prediction in smart cities. Information Sciences 542, 476–491

    work page 2021

  8. [6]

    Microestimates of wealth for all low-and middle-income countries

    Chi, G., Fang, H., Chatterjee, S., Blumenstock, J.E., 2022. Microestimates of wealth for all low-and middle-income countries. Proceedings of the National Academy of Sciences 119, e2113658119

    work page 2022

Show all 40 references
  1. [7]

    Social vulnerability to environmental hazards

    Cutter, S.L., Boruff, B.J., Shirley, W.L., 2003. Social vulnerability to environmental hazards. Social science quarterly 84, 242–261

    2003

  2. [8]

    Association betweensocialvulnerabilityandacounty’sriskforbecomingacovid-19hotspot—unitedstates,june1–july25,2020

    Dasgupta,S.,Bowen,V.B.,Leidner,A.,Fletcher,K.,Musial,T.,Rose,C.,Cha,A.,Kang,G.,Dirlikov,E.,Pevzner,E.,etal.,2020. Association betweensocialvulnerabilityandacounty’sriskforbecomingacovid-19hotspot—unitedstates,june1–july25,2020. MorbidityandMortality Weekly Report 69, 1535

    2020

  3. [9]

    Time-seriesrepresentationlearningviatemporalandcontextual contrasting, in: International Joint Conference on Artificial Intelligence

    Eldele,E.,Ragab,M.,Chen,Z.,Wu,M.,Kwoh,C.K.,Li,X.,Guan,C.,2021. Time-seriesrepresentationlearningviatemporalandcontextual contrasting, in: International Joint Conference on Artificial Intelligence

    2021

  4. [10]

    A social vulnerability index for disaster management

    Flanagan, B.E., Gregory, E.W., Hallisey, E.J., Heitgerd, J.L., Lewis, B., 2011. A social vulnerability index for disaster management. Journal of homeland security and emergency management 8, 0000102202154773551792

    2011

  5. [11]

    Unsupervised scalable representation learning for multivariate time series

    Franceschi, J.Y., Dieuleveut, A., Jaggi, M., 2019. Unsupervised scalable representation learning for multivariate time series. Advances in neural information processing systems 32

    2019

  6. [12]

    Fu, Y., Wang, P., Du, J., Wu, L., Li, X., 2019. Efficient region embedding with multi-view spatial networks: A perspective of locality- constrained spatial autocorrelations, in: Proceedings of the AAAI Conference on Artificial Intelligence, pp. 906–913

    2019

  7. [13]

    Semantictrajectoryrepresentationandretrievalviahierarchicalembedding

    Gao,C.,Zhang,Z.,Huang,C.,Yin,H.,Yang,Q.,Shao,J.,2020. Semantictrajectoryrepresentationandretrievalviahierarchicalembedding. Information Sciences 538, 176–192

    2020

  8. [14]

    Huang, T., Wang, Z., Sheng, H., Ng, A.Y., Rajagopal, R., 2021. M3g: Learning urban neighborhood representation from multi-modal multi- graph, in: Proceedings of the DeepSpatial 2021: 2nd ACM KDD Workshop on Deep Learning for Spatio-Temporal Data, Applications and Systems

    2021

  9. [15]

    Effective urban region representation learning using heterogeneous urban graph attention network (hugat)

    Kim, N., Yoon, Y., 2022. Effective urban region representation learning using heterogeneous urban graph attention network (hugat). arXiv preprint arXiv:2202.09021

    2022 arXiv

  10. [16]

    Urban region representation learning with openstreetmap building footprints, in: Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pp

    Li, Y., Huang, W., Cong, G., Wang, H., Wang, Z., 2023. Urban region representation learning with openstreetmap building footprints, in: Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pp. 1363–1373

    2023

  11. [18]

    Revealingspatio-temporalevolutionofurbanvisualenvironmentswithstreetviewimagery

    Liang,X.,Zhao,T.,Biljecki,F.,2023. Revealingspatio-temporalevolutionofurbanvisualenvironmentswithstreetviewimagery. Landscape and Urban Planning 237, 104802

    2023

  12. [19]

    Exploringtrajectoryembeddingviaspatial-temporalpropagationfordynamicregion representations

    Liu,C.,Zhang,H.,Zhu,G.,Guan,H.,Kwong,S.,2024. Exploringtrajectoryembeddingviaspatial-temporalpropagationfordynamicregion representations. Information Sciences 668, 120516

    2024

  13. [20]

    Urban flow pattern mining based on multi-source heterogeneous data fusion and knowledge graph embedding

    Liu, J., Li, T., Ji, S., Xie, P., Du, S., Teng, F., Zhang, J., 2021. Urban flow pattern mining based on multi-source heterogeneous data fusion and knowledge graph embedding. IEEE Transactions on Knowledge and Data Engineering 35, 2133–2146

    2021

  14. [21]

    Mhccl: Masked hierarchical cluster-wise contrastive learning for multivariate time series, in: Proceedings of the AAAI Conference on Artificial Intelligence, pp

    Meng, Q., Qian, H., Liu, Y., Cui, L., Xu, Y., Shen, Z., 2023. Mhccl: Masked hierarchical cluster-wise contrastive learning for multivariate time series, in: Proceedings of the AAAI Conference on Artificial Intelligence, pp. 9153–9161

    2023

  15. [22]

    Wavenet: A generative model for raw audio

    Oord, A.v.d., Dieleman, S., Zen, H., Simonyan, K., Vinyals, O., Graves, A., Kalchbrenner, N., Senior, A., Kavukcuoglu, K., 2016. Wavenet: A generative model for raw audio. arXiv preprint arXiv:1609.03499

    2016 arXiv

  16. [23]

    Representation learning with contrastive predictive coding

    Oord, A.v.d., Li, Y., Vinyals, O., 2018. Representation learning with contrastive predictive coding. arXiv preprint arXiv:1807.03748

    2018 arXiv

  17. [24]

    Pytorch:An imperative style, high-performance deep learning library

    Paszke,A.,Gross,S.,Massa,F.,Lerer,A.,Bradbury,J.,Chanan,G.,Killeen,T.,Lin,Z.,Gimelshein,N.,Antiga,L.,etal.,2019. Pytorch:An imperative style, high-performance deep learning library. Advances in neural information processing systems 32

    2019

  18. [25]

    Improveddeepmetriclearningwithmulti-classn-pairlossobjective

    Sohn,K.,2016. Improveddeepmetriclearningwithmulti-classn-pairlossobjective. Advancesinneuralinformationprocessingsystems29

    2016

  19. [26]

    Most common modes of transportation for commuting in the u.s

    Statista, 2024. Most common modes of transportation for commuting in the u.s. as of december 2023.https://www.statista.com/ forecasts/997176/most-common-modes-of-transportation-for-commuting-in-the-us

    2024

  20. [27]

    Unsupervised representation learning for time series with temporal neighborhood coding, in: International Conference on Learning Representations

    Tonekaboni, S., Eytan, D., Goldenberg, A., 2021. Unsupervised representation learning for time series with temporal neighborhood coding, in: International Conference on Learning Representations

    2021

  21. [28]

    Graph attention networks

    Veličković, P., Cucurull, G., Casanova, A., Romero, A., Liò, P., Bengio, Y., 2017. Graph attention networks. International Conference on Learning Representations

    2017

  22. [29]

    Region representation learning via mobility flow, in: Proceedings of the 2017 ACM on Conference on Information and Knowledge Management, pp

    Wang, H., Li, Z., 2017. Region representation learning via mobility flow, in: Proceedings of the 2017 ACM on Conference on Information and Knowledge Management, pp. 237–246

    2017

  23. [30]

    Mixingupcontrastivelearning:Self-supervisedrepresentationlearning for time series

    Wickstrøm,K.,Kampffmeyer,M.,Mikalsen,K.Ø.,Jenssen,R.,2022. Mixingupcontrastivelearning:Self-supervisedrepresentationlearning for time series. Pattern Recognition Letters 155, 54–61

    2022

  24. [31]

    Multi-graph fusion networks for urban region embedding

    Wu, S., Yan, X., Fan, X., Pan, S., Zhu, S., Zheng, C., Cheng, M., Wang, C., 2022. Multi-graph fusion networks for urban region embedding. arXiv preprint arXiv:2201.09760

    2022 arXiv

  25. [32]

    Unsupervised feature learning via non-parametric instance discrimination, in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp

    Wu, Z., Xiong, Y., Yu, S.X., Lin, D., 2018. Unsupervised feature learning via non-parametric instance discrimination, in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 3733–3742. N Kim et al.: Preprint submitted to Elsevier Page 12 of 13 Mo...

    2018

  26. [33]

    Assessing dynamics of human vulnerability at community level–using mobility data

    Xia, C., Hu, Y., Chi, G., Chen, J., 2023. Assessing dynamics of human vulnerability at community level–using mobility data. International Journal of Disaster Risk Reduction 96, 103964

    2023

  27. [34]

    Yao, Z., Fu, Y., Liu, B., Hu, W., Xiong, H., 2018. Representing urban functions through zone embedding with human mobility patterns, in: Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence (IJCAI-18)

    2018

  28. [35]

    Yu, F., Koltun, V., 2016. Multi-scale context aggregation by dilated convolutions, in: International Conference on Learning Representations, San Juan, Puerto Rico, May 2-4, 2016, Conference Track Proceedings

    2016

  29. [36]

    Ts2vec: Towards universal representation of time series, in: AAAI Conference on Artificial Intelligence

    Yue, Z., Wang, Y., Duan, J., Yang, T., Huang, C., Tong, Y., Xu, B., 2022. Ts2vec: Towards universal representation of time series, in: AAAI Conference on Artificial Intelligence

    2022

  30. [38]

    Zhang, M., Li, T., Li, Y., Hui, P., 2021a. Multi-view joint graph representation learning for urban region embedding, in: Proceedings of the twenty-ninth international conference on international joint conferences on artificial intelligence, pp. 4431–4437

  31. [39]

    Unveiling transit mobility structure towards sustainable cities: An integrated graph embedding approach

    Zhang, T., Duan, X., Li, Y., 2021b. Unveiling transit mobility structure towards sustainable cities: An integrated graph embedding approach. Sustainable Cities and Society 72, 103027

  32. [40]

    Learning region similarities via graph-based deep metric learning

    Zhao, Y., Qi, J., Trisedya, B.D., Su, Y., Zhang, R., Ren, H., 2023. Learning region similarities via graph-based deep metric learning. IEEE Transactions on Knowledge and Data Engineering 35, 10237–10250. N Kim et al.: Preprint submitted to Elsevier Page 13 of 13

    2023

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.