Pith. sign in

REVIEW 3 major objections 6 minor 46 references

Trajectory Representation Learning on Road Networks and Grids with Spatio-Temporal Dynamics

T0 review · 3 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read TIGR fuses grid, road, and traffic views into one trajectory embedding that outperforms single-modality baselines on every downstream task, with gains up to 43.22% in similarity, 16.65% in travel time, and 10.16% in destination prediction.

desk verdict TIGR's three-branch architecture and grid-vs-road comparison are worth engaging, but the missing train/test split and possible leakage from traffic statistics make the headline results unverified. read the letter →

arxiv 2411.14014 v2 pith:M2QRCQ24 submitted 2024-11-21 cs.LG cs.AI

Stefan Schestakov , Simon Gottschalk This is my paper · ORCID
classification cs.LGcs.AI
keywords trajectoryrepresentationlearningspatio-temporaldynamicsgridandroadmodalitiescontrastivesimilaritytraveltimeestimationdestinationpredictionmulti-modalfusion
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 argues that trajectory representation learning has been unnecessarily split into grid-based and road-network-based camps, and that both camps ignore the fact that traffic changes over time. TIGR processes a trajectory through three parallel branches—grid cells, road segments, and a spatio-temporal branch carrying hourly traffic speeds and road-transition probabilities—and aligns the three views with contrastive learning. The claim is that the resulting single embedding is a general-purpose trajectory representation: on two real taxi datasets it beats every single-modality baseline on trajectory similarity, travel time estimation, and destination prediction. If true, one pretrained embedding can replace task-specific models, and the modal split in the field is a real loss.

What carries the argument

The load-bearing object is a three-branch encoder with a shared contrastive training objective. Each branch embeds a trajectory into tokens: grid cells via node2vec-initialized embeddings, road segments via node2vec-initialized embeddings, and a spatio-temporal branch that fuses a dynamic traffic embedding—computed by graph convolution over a transition-probability matrix $P$ (Equation 1) weighted by hourly mean traffic speeds $X$—with a learnable cosine temporal embedding (Equation 3). The fusion uses local multi-head attention (LMA), which splits the sequence into $H$ subsequences and applies attention within each head, forcing attention to stay local in space and time. Two masked views per branch are encoded by a Transformer with RoPE and RMSNorm, with a target encoder updated by exponential moving average, and InfoNCE losses align views within each branch (intra-modal) and across the grid/road and road/spatio-temporal pairs (inter-modal). The final representation is the concatenation of the three branch embeddings.

What would settle it

Re-run the Porto and San Francisco experiments with the transition probabilities and hourly speeds computed strictly from the training split (e.g., a temporal split where all test trips happen after all training trips), and check whether TIGR's margin over TrajCL and START persists; if the margin collapses, the central claim of general-purpose superiority is an artifact of information leakage. A cheaper check: remove the spatio-temporal branch entirely (Grid+Road) and see on a third city how much of the reported gain remains.

Watch

Extended reading notes

Core claim

TIGR's central claim is that integrating grid and road network modalities with spatio-temporal dynamics produces trajectory embeddings that are strictly better than any single-modality embedding on all three downstream tasks tested. Specifically, on the Porto and San Francisco taxi datasets, TIGR reports mean-rank improvements up to 43.22% for trajectory similarity, MAPE improvements up to 16.65% for travel time estimation, and F1/accuracy improvements up to 10.16% for destination prediction, beating nine baselines (t2vec, CLT-Sim, TrajCL, CSTTE, Trembr, Toast, JCLRNT, LightPath, START) and a vanilla Transformer baseline on every metric. The paper also establishes a comparative result: with identical architecture, road-based representations are better for similarity search while grid-based representations are better for travel time, and the spatio-temporal branch alone is weak on destination prediction but strong when fused with structural branches.

Load-bearing premise

The evaluation assumes that the traffic statistics used in the model—how often one road leads to another and the average speed by hour—are built only from training trips and not from the trips used for testing, but the paper never states this split.

Editorial extensions

If this is right

  • One TIGR embedding, frozen after pretraining, powers trajectory similarity, travel time estimation, and destination prediction; the similarity results are near-perfect (HR@1 of 0.976 on Porto and 0.952 on San Francisco).
  • Grid and road information are complementary in a measurable way: the grid branch alone gives 41.5% better MAPE than the road branch on Porto, while the road branch gives 4.6% higher HR@1 and 8.8% higher Acc@1.
  • Adding the spatio-temporal branch to either structural branch helps: it improves HR@1 by 23.4% and MAPE by 18.8% when added to the road branch on Porto.
  • Masking strategy is a first-order choice: truncation plus consecutive masking outperforms single-strategy masking, with a 25.4% swing in HR@1 between the worst and best combinations.
  • TIGR's similarity retrieval degrades more slowly than baselines as the negative-sample pool grows to 100k, suggesting the fused embedding separates similar from dissimilar trajectories more robustly.

Reading between the lines

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

  • A direct test of the leakage concern: recompute the transition probability matrix $P$ and hourly speed matrix $X$ from the training split only, re-run the pipeline, and check whether the 43% similarity gain survives; the paper never specifies this split.
  • The reported modality asymmetry suggests a cheap ensemble baseline: concatenate a strong grid-model embedding (e.g., TrajCL) with a strong road-model embedding (e.g., START) and train the same MLP heads; the paper's Grid+Road ablation (Acc@1 0.241 on Porto) is the natural comparison, but a direct external evaluation would clarify how much the spatio-temporal branch adds beyond simple concatenation
  • The inverse correlation between trajectory-similarity and destination-prediction performance across mask ratios and embedding dimensions implies a Pareto frontier: a single 512-dimension embedding balances them, but task-specific embeddings chosen from the frontier could do better on each task.
  • Both datasets are taxi fleets in single cities; the spatio-temporal branch depends on hourly speed aggregates that may be far noisier for pedestrian, bike, or ride-pooling data, so the 43% margin is an upper bound until tested on other mobility modes.
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

3 major / 6 minor

Summary. This paper proposes TIGR, a self-supervised trajectory representation learning model with three parallel branches: a grid-cell branch, a road-network branch, and a spatio-temporal branch built from transition-probability graph convolutions, temporal embeddings, and local multi-head attention. Two masked views of each trajectory are aligned within each branch and across branches using InfoNCE losses, and the final representation is a concatenation of the three branch encodings. The model is evaluated on Porto and San Francisco for trajectory similarity, travel time estimation, and destination prediction, with reported improvements of up to 43.22%, 16.65%, and 10.16% over baselines, together with ablations and a comparison of grid-based versus road-based methods.

Significance. If the reported results are reproducible under a clean evaluation protocol, the paper makes a useful empirical contribution: it provides the first systematic comparison of grid-based and road-based trajectory representation learning, demonstrates complementary strengths of the two modalities, and introduces a plausible three-branch fusion that incorporates dynamic traffic statistics. The ablation study in Table II is internally consistent, and the release of the training and evaluation framework is a practical asset for the community. However, the central empirical claim is currently weakened by an underspecified train/test protocol and by hyperparameter selection on the test dataset; the significance of the reported gains depends on ruling out information leakage in the traffic statistics and on validating the design choices on a separate split.

major comments (3)
  1. [Section III-A1 / IV-A / IV-C] The evaluation never defines a split into training, validation, and test trajectories. The spatio-temporal branch in Section III-A1 constructs the transition probability matrix P from 'historical trajectories' and the hourly traffic-speed matrix X from aggregated traffic; if these statistics are computed from the full Porto or San Francisco datasets, then every test trajectory contributes to the traffic statistics used to build its own representation. This would directly inflate the travel-time results in Table I, because X contains traffic states for the very trips whose durations are predicted, and could also make the trajectory-similarity retrieval in Section IV-C easier. Please specify exactly which trajectories are used to compute P and X, which trajectories form the pretraining corpus, and how each downstream task is split into training and test sets, and rerun the experiments with strict disjoint splits.
  2. [Section V-D / V-F] The masking strategies and hyperparameters are selected using downstream performance on the Porto dataset: Section V-D chooses the masking combination from Figure 5 on the TS task, and Section V-F chooses encoder depth, queue size, embedding dimension, and masking ratios from Figure 6 across the Porto tasks. Table I then reports Porto performance for this selected configuration. Without a separate validation split, the Porto numbers are optimized on the evaluation set, and the reported gains over baselines may include selection bias. Please introduce a validation partition for all hyperparameter and masking choices and report test performance on a held-out split.
  3. [Section IV-C / III-A1] For travel time estimation, Section IV-C states that the model 'only feed[s] the start time and mask[s] all other timestamps,' but it is not described how this masking is applied to each branch. The spatio-temporal branch uses hourly traffic states xth for each road segment and temporal embeddings ti for every point (Equations 2 and 3). If the frozen encoder still receives the full sequence of timestamps or the traffic states for hours after the start time, the travel time label is contained in the input. Please clarify the exact input representation used for TTE, confirm that the completion time and all post-start traffic states are hidden from every branch, and report whether the same masking is applied to baselines.
minor comments (6)
  1. [Section III-A1, Eq. (2)] The notation x(td,th)_i in Equation (2) is inconsistent with the definition of X as a function of hour th only, and the symbol td is never defined; please correct the indexing.
  2. [Figure 3] The legend label 'JCML' should be 'JCLRNT' to match the baseline name used elsewhere in the paper.
  3. [Figure 6] The y-axis label 'TS HR@1 (%)' with values between 0.94 and 0.98 is inconsistent; use either fractions or percentages consistently.
  4. [Section V-C] The claim that adding the spatio-temporal branch to the road branch improves HR@1 by 23.4% does not match Table II: Road alone has HR@1 0.734 and Road+ST has 0.958, which is a relative improvement of about 30.5%; please verify the reported percentage.
  5. [Section III-C] The notation for the two masked views, printed as 'bTb' and 'eTb', is difficult to read and is not defined in the text; please use clear symbols such as hat T and tilde T and define them at first use.
  6. [References] Reference [2] lists an author as 'J. H. Additional,' which appears to be a placeholder; please correct the author list.

Circularity Check

0 steps flagged · score 2.0 of 10

No definitional circularity: TIGR's central claim is an empirical benchmark comparison with a self-contained self-supervised objective; the only self-citation is non-load-bearing, though the evaluation omits a stated train/test split for traffic statistics (a leakage risk rather than a circularity).

full rationale

TIGR's derivation chain is not circular by construction. The trajectory encoder is trained with self-supervised intra- and inter-modal InfoNCE losses (Eqs. 8-11), and the downstream tasks only train an MLP on frozen embeddings (Section IV-C), so the downstream predictions are not fitted parameters renamed as predictions. The spatio-temporal branch uses transition probabilities P (Eq. 1) and hourly mean speeds X computed from 'historical trajectories' (Section III-A1); if these statistics are computed only from the training split, the evaluation is legitimate, and the paper's equations do not define P or X in terms of the downstream travel-time label. The only self-citation, reference [38], appears in Related Work and is not load-bearing. The manuscript does omit an explicit train/test split for the aggregate traffic statistics in Section IV-A, and for TTE the dynamic traffic embedding at the start hour uses X; if X includes test trips, the reported up-to-16.65% TTE improvement and the TS results could be inflated by leakage. However, this is a protocol-level correctness risk, not an equivalence-by-construction or a self-citation chain. The near-perfect TS numbers may also reflect the odd/even-query construction, but again this is task design, not circularity. Overall, the central empirical claim has independent content and is benchmarked against external datasets and baselines, so no significant circularity is established.

Assumptions & free parameters 7 free parameters · 7 assumptions · 0 invented entities

TIGR is an empirical ML system and does not introduce physical entities. The central result rests on preprocessing choices, traffic-statistic construction, and hyperparameter selection, all listed above. The most consequential assumption is that P and X are computed from training trajectories only, which is not stated explicitly. The most consequential selection choice is tuning masking and architecture hyperparameters on the Porto downstream evaluation without a described validation split.

free parameters (7)
  • Masking ratios pRM, pTC, pCM = 0.3
    Set to 0.3 based on parameter analysis on Porto downstream tasks (Section V-F); the final model uses TC and CM for View 1 and all three strategies for View 2 (Section V-D).
  • Loss weight lambda = 0.5
    Chosen in Section IV-D to balance inter- and intra-modal losses; no sensitivity analysis is shown for lambda.
  • Temperature tau = 0.05
    Set in Section IV-D for the InfoNCE loss; no sensitivity analysis is reported.
  • EMA target decay rate mu = not reported
    Defined in Section III-C as the target decay rate for the EMA update, but its value is never specified.
  • Encoder depth, queue size, embedding dimensions = 2 layers, queue 2048, 512 total dim (grid 256, road and ST 128)
    Selected from the parameter study on Porto (Section V-F), including the observation that TS peaks at 2 layers and dimension 512 balances TTE and DP.
  • LMA number of heads H = not reported
    Local multi-head attention in Eq. 6 depends on the number of heads H, but the paper does not state H.
  • Transition count smoothing constant = 1
    The +1 in Eq. 1 is a Laplace-style smoothing choice; its effect is not analyzed.
assumptions (7)
  • standard math Transformer, RoPE, RMSNorm, InfoNCE, and EMA are assumed to work as described in the cited literature.
    Invoked in Sections III-C and III-D as the backbone of the encoders and contrastive losses.
  • domain assumption Fast Map Matching correctly maps GPS points to the correct road segments.
    Section IV-A uses Fast Map Matching [18] to create road trajectories; errors in map matching propagate into the road and spatio-temporal branches.
  • domain assumption The road network is static and its topology is complete for both cities.
    Road networks extracted from OpenStreetMap (Section IV-A) are treated as a fixed graph G in Definition 4; missing or outdated roads would bias both road and spatio-temporal branches.
  • domain assumption Hourly aggregated traffic speeds X and transition probabilities P, computed from historical trajectories, are representative and stationary across the evaluation period.
    Section III-A1 defines X and P from trajectory data; if traffic distributions shift or if statistics include test trips, the embeddings and downstream estimates are not fairly evaluated.
  • ad hoc to paper Traffic within a small spatial and temporal window is similar, motivating local multi-head attention.
    This heuristic justifies LMA in Section III-A3; it is not derived or independently validated.
  • ad hoc to paper Removing grid-ST inter-modal alignment is justified by empirical degradation.
    Section III-D states that aligning grid and spatio-temporal branches 'empirically degraded' results, an ad hoc design choice based on observed performance.
  • domain assumption Pruning trajectories with fewer than 20 or more than 200 points does not bias downstream tasks.
    Section IV-A applies these filters without analysis of how the filtering changes the test distribution.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Trajectory Representation Learning on Road Networks and Grids with Spatio-Temporal Dynamics." pith.science (2026). https://pith.science/paper/M2QRCQ24

@misc{pith2026241114014,
  author       = {Pith},
  title        = {Pith review of: Trajectory Representation Learning on Road Networks and Grids with Spatio-Temporal Dynamics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/M2QRCQ24}},
  note         = {Machine review of arXiv:2411.14014}
}
read the original abstract

Trajectory representation learning is a fundamental task for applications in fields including smart city, and urban planning, as it facilitates the utilization of trajectory data (e.g., vehicle movements) for various downstream applications, such as trajectory similarity computation or travel time estimation. This is achieved by learning low-dimensional representations from high-dimensional and raw trajectory data. However, existing methods for trajectory representation learning either rely on grid-based or road-based representations, which are inherently different and thus, could lose information contained in the other modality. Moreover, these methods overlook the dynamic nature of urban traffic, relying on static road network features rather than time varying traffic patterns. In this paper, we propose TIGR, a novel model designed to integrate grid and road network modalities while incorporating spatio-temporal dynamics to learn rich, general-purpose representations of trajectories. We evaluate TIGR on two realworld datasets and demonstrate the effectiveness of combining both modalities by substantially outperforming state-of-the-art methods, i.e., up to 43.22% for trajectory similarity, up to 16.65% for travel time estimation, and up to 10.16% for destination prediction.

Figures

Figures reproduced from arXiv: 2411.14014 by the authors.

Figure 1
Figure 1. Trajectory represented by grid cells (a) and by road segments on a [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Framework of TIGR’s training procedure. 2) Temporal Embedding: Inspired by the sinusoidal Posi￾tional Embeddings [20], we learn time embeddings by utilizing the cosine function to capture periodic behavior. Let ti be the i￾th time point in a trajectory, then we obtain its time embedding ti : ti [k] = ( wkti + ϕk if k = 0, cos(wkti + ϕk) else, (3) where ti [k] is the k-th element of the embedding vector ti ∈ R q with… view at source ↗
Figure 3
Figure 3. Performance on TS (HR@1 ↑) with increasing amount of negative samples kneg. TABLE II ABLATION STUDY ON THE PORTO DATASET. TS TTE DP HR@1 ↑ MAPE ↓ Acc@1 ↑ w/o Road & ST (= only Grid) 0.702 13.49 0.193 w/o Grid & ST (= only Road) 0.734 23.05 0.210 w/o Road & Grid (= only ST) 0.596 20.09 0.031 w/o ST branch (= Grid & Road) 0.849 13.33 0.241 w/o Road branch (= Grid & ST) 0.931 13.06 0.189 w/o Grid branch (= Road & ST) 0… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Given the query trajectory (red), we visualize the 500 most similar trajectories retrieved by [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Impact of the masking strategies on the Porto dataset. We report [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Parameter analysis on the Porto dataset. [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 40 canonical work pages

  1. [1]

    In- terpreting trajectories from multiple views: A hierarchical self-attention network for estimating the time of arrival,

    Z. Chen, X. Xiao, Y . Gong, J. Fang, N. Ma, H. Chai, and Z. Cao, “In- terpreting trajectories from multiple views: A hierarchical self-attention network for estimating the time of arrival,” in 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD . ACM, 2022, pp. 2771–2779

    work page 2022

  2. [2]

    Dual graph convolution architecture search for travel time estimation,

    G. Jin, H. Yan, F. Li, Y . Li, J. H. Additional, and J. Huang, “Dual graph convolution architecture search for travel time estimation,” ACM Transactions on Intelligent Systems and Technology , 2023

    work page 2023

  3. [3]

    Travel time distribution estimation by learning representations over temporal attributed graphs,

    W. Zhou, X. Xiao, Y . Gong, J. Chen, J. Fang, N. Tan, N. Ma, Q. Li, H. Chai, S. Jeon, and J. Zhang, “Travel time distribution estimation by learning representations over temporal attributed graphs,” IEEE Trans. Intell. Transp. Syst. , vol. 24, no. 5, pp. 5069–5081, 2023

    work page 2023

  4. [4]

    Destination prediction by trajectory distribution-based model,

    P. C. Besse, B. Guillouet, J. Loubes, and F. Royer, “Destination prediction by trajectory distribution-based model,” IEEE Trans. Intell. Transp. Syst., vol. 19, no. 8, pp. 2470–2481, 2018

    work page 2018

  5. [5]

    Destination prediction based on virtual POI docks in dockless bike-sharing system,

    M. Jiang, C. Li, K. Li, and H. Liu, “Destination prediction based on virtual POI docks in dockless bike-sharing system,” IEEE Trans. Intell. Transp. Syst., vol. 23, no. 3, pp. 2457–2470, 2022

    work page 2022

  6. [6]

    How machine learning informs ride-hailing services: A survey,

    Y . Liu, R. Jia, J. Ye, and X. Qu, “How machine learning informs ride-hailing services: A survey,” Communications in Transportation Research, vol. 2, p. 100075, 2022

    2022

  7. [7]

    Spatio-temporal trajectory similarity learning in road networks,

    Z. Fang, Y . Du, X. Zhu, D. Hu, L. Chen, Y . Gao, and C. S. Jensen, “Spatio-temporal trajectory similarity learning in road networks,” in 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD. ACM, 2022, pp. 347–356

    work page 2022

  8. [8]

    Trajgat: A graph- based long-term dependency modeling approach for trajectory similarity computation,

    D. Yao, H. Hu, L. Du, G. Cong, S. Han, and J. Bi, “Trajgat: A graph- based long-term dependency modeling approach for trajectory similarity computation,” in 28th ACM SIGKDD Conference on Knowledge Discov- ery and Data Mining, KDD . ACM, 2022, pp. 2275–2285

    work page 2022

Show all 46 references
  1. [9]

    KGTS: contrastive trajectory similarity learning over prompt knowl- edge graph embedding,

    Z. Chen, D. Zhang, S. Feng, K. Chen, L. Chen, P. Han, and S. Shang, “KGTS: contrastive trajectory similarity learning over prompt knowl- edge graph embedding,” in Thirty-Eighth Conference on Artificial Intel- ligence, AAAI, 2024

    2024

  2. [10]

    Contrastive trajectory simi- larity learning with dual-feature attention,

    Y . Chang, J. Qi, Y . Liang, and E. Tanin, “Contrastive trajectory simi- larity learning with dual-feature attention,” in 39th IEEE International Conference on Data Engineering, ICDE 2023 . IEEE, 2023, pp. 2933– 2945

    2023

  3. [11]

    Contrastive pre-training of spatial- temporal trajectory embeddings,

    Y . Lin, H. Wan, S. Guo, and Y . Lin, “Contrastive pre-training of spatial- temporal trajectory embeddings,” CoRR, vol. abs/2207.14539, 2022

    2022 arXiv

  4. [12]

    Efficient trajectory similarity computation with contrastive learning,

    L. Deng, Y . Zhao, Z. Fu, H. Sun, S. Liu, and K. Zheng, “Efficient trajectory similarity computation with contrastive learning,” in 31st ACM International Conference on Information & Knowledge Management, CIKM. ACM, 2022, pp. 365–374

    2022

  5. [13]

    Deep representation learning for trajectory similarity computation,

    X. Li, K. Zhao, G. Cong, C. S. Jensen, and W. Wei, “Deep representation learning for trajectory similarity computation,” in 34th IEEE Interna- tional Conference on Data Engineering, ICDE 2018 . IEEE Computer Society, 2018, pp. 617–628

    2018

  6. [14]

    Self-supervised trajectory representation learning with temporal regularities and travel semantics,

    J. Jiang, D. Pan, H. Ren, X. Jiang, C. Li, and J. Wang, “Self-supervised trajectory representation learning with temporal regularities and travel semantics,” in 39th IEEE International Conference on Data Engineering, ICDE 2023. IEEE, 2023, pp. 843–855

    2023

  7. [15]

    Lightpath: Lightweight and scalable path representation learning,

    S. B. Yang, J. Hu, C. Guo, B. Yang, and C. S. Jensen, “Lightpath: Lightweight and scalable path representation learning,” in 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD

  8. [16]

    Jointly contrastive representation learning on road network and trajectory,

    Z. Mao, Z. Li, D. Li, L. Bai, and R. Zhao, “Jointly contrastive representation learning on road network and trajectory,” in 31st ACM International Conference on Information & Knowledge Management, CIKM. ACM, 2022, pp. 1501–1510

    2022

  9. [17]

    Trembr: Exploring road networks for trajectory representation learning,

    T. Fu and W. Lee, “Trembr: Exploring road networks for trajectory representation learning,” ACM Trans. Intell. Syst. Technol., vol. 11, no. 1, 2020

    2020

  10. [18]

    Fast map matching, an algorithm integrating hidden markov model with precomputation,

    C. Yang and G. Gid ´ofalvi, “Fast map matching, an algorithm integrating hidden markov model with precomputation,” Int. J. Geogr. Inf. Sci. , vol. 32, no. 3, pp. 547–570, 2018

    2018

  11. [19]

    Pre-training general trajectory embeddings with maximum multi-view entropy cod- ing,

    Y . Lin, H. Wan, S. Guo, J. Hu, C. S. Jensen, and Y . Lin, “Pre-training general trajectory embeddings with maximum multi-view entropy cod- ing,” IEEE Transactions on Knowledge and Data Engineering , 2023

    2023

  12. [20]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention is all you need,” in Annual Conference on Neural Information Processing Systems, NeurIPS , 2017, pp. 5998–6008

    2017

  13. [21]

    Masked autoencoders are scalable vision learners,

    K. He, X. Chen, S. Xie, Y . Li, P. Doll ´ar, and R. B. Girshick, “Masked autoencoders are scalable vision learners,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022 . IEEE, 2022, pp. 15 979–15 988

    2022

  14. [22]

    Masked autoencoders as spatiotemporal learners,

    C. Feichtenhofer, H. Fan, Y . Li, and K. He, “Masked autoencoders as spatiotemporal learners,” in Annual Conference on Neural Information Processing Systems, NeurIPS , 2022

    2022

  15. [23]

    Mavil: Masked audio-video learners,

    P. Huang, V . Sharma, H. Xu, C. Ryali, H. Fan, Y . Li, S. Li, G. Ghosh, J. Malik, and C. Feichtenhofer, “Mavil: Masked audio-video learners,” in Annual Conference on Neural Information Processing Systems, NeurIPS, 2023

    2023

  16. [24]

    BERT: pre-training of deep bidirectional transformers for language understanding,

    J. Devlin, M. Chang, K. Lee, and K. Toutanova, “BERT: pre-training of deep bidirectional transformers for language understanding,” in NAACL- HLT. Association for Computational Linguistics, 2019, pp. 4171–4186

    2019

  17. [25]

    Masked siamese networks for label-efficient learning,

    M. Assran, M. Caron, I. Misra, P. Bojanowski, F. Bordes, P. Vincent, A. Joulin, M. Rabbat, and N. Ballas, “Masked siamese networks for label-efficient learning,” in 17th European Conference on Computer Vision ECCV, 2022

    2022

  18. [26]

    Roformer: Enhanced trans- former with rotary position embedding,

    J. Su, Y . Lu, S. Pan, B. Wen, and Y . Liu, “Roformer: Enhanced trans- former with rotary position embedding,” CoRR, vol. abs/2104.09864, 2021

    2021 arXiv

  19. [27]

    Root mean square layer normalization,

    B. Zhang and R. Sennrich, “Root mean square layer normalization,” in Annual Conference on Neural Information Processing Systems, NeurIPS, 2019, pp. 12 360–12 371

    2019

  20. [28]

    With a little help from my friends: Nearest-neighbor contrastive learning of visual representations,

    D. Dwibedi, Y . Aytar, J. Tompson, P. Sermanet, and A. Zisserman, “With a little help from my friends: Nearest-neighbor contrastive learning of visual representations,” in IEEE/CVF International Conference on Computer Vision, 2021, pp. 9588–9597

    2021

  21. [29]

    Representation learning with contrastive predictive coding,

    A. Oord et al. , “Representation learning with contrastive predictive coding,” CoRR, vol. abs/1807.03748, 2018

    2018 arXiv

  22. [30]

    Trajectory similarity learning with auxiliary supervision and optimal matching,

    H. Zhang, X. Zhang, Q. Jiang, B. Zheng, Z. Sun, W. Sun, and C. Wang, “Trajectory similarity learning with auxiliary supervision and optimal matching,” in 29th International Joint Conference on Artificial Intelligence, IJCAI. ijcai.org, 2020, pp. 3209–3215

    2020

  23. [31]

    T3S: effective representation learning for trajectory similarity computation,

    P. Yang, H. Wang, Y . Zhang, L. Qin, W. Zhang, and X. Lin, “T3S: effective representation learning for trajectory similarity computation,” in 37th IEEE International Conference on Data Engineering, ICDE . IEEE, 2021, pp. 2183–2188

    2021

  24. [32]

    Weakly- supervised temporal path representation learning with contrastive cur- riculum learning,

    S. B. Yang, C. Guo, J. Hu, B. Yang, J. Tang, and C. S. Jensen, “Weakly- supervised temporal path representation learning with contrastive cur- riculum learning,” in 38th IEEE International Conference on Data Engineering, ICDE. IEEE, 2022

    2022

  25. [33]

    Robust road network representation learning: When traffic patterns meet traveling semantics,

    Y . Chen, X. Li, G. Cong, Z. Bao, C. Long, Y . Liu, A. K. Chandran, and R. Ellison, “Robust road network representation learning: When traffic patterns meet traveling semantics,” in ACM International Conference on Information and Knowledge Management, CIKM . ACM, 2021, pp. 211–220

    2021

  26. [34]

    Deepwalk: online learning of social representations,

    B. Perozzi, R. Al-Rfou, and S. Skiena, “Deepwalk: online learning of social representations,” in ACM SIGKDD. ACM, 2014, pp. 701–710

    2014

  27. [35]

    Bootstrap your own latent - A new approach to self-supervised learning,

    J. Grill, F. Strub, F. Altch ´e, C. Tallec, P. H. Richemond, E. Buchatskaya, C. Doersch, B. ´A. Pires, Z. Guo, M. G. Azar, B. Piot, K. Kavukcuoglu, R. Munos, and M. Valko, “Bootstrap your own latent - A new approach to self-supervised learning,” in Annual Conference on Neural ...

    2020

  28. [36]

    A simple framework for contrastive learning of visual representations,

    T. Chen, S. Kornblith, M. Norouzi, and G. Hinton, “A simple framework for contrastive learning of visual representations,” in International Conference on Machine Learning, ICML . PMLR, 2020, pp. 1597– 1607

    2020

  29. [37]

    node2vec: Scalable feature learning for networks,

    A. Grover and J. Leskovec, “node2vec: Scalable feature learning for networks,” in KDD, 2016

    2016

  30. [38]

    Metacitta: Deep meta-learning for spatio-temporal prediction across cities and tasks,

    A. Sao, S. Gottschalk, N. Tempelmeier, and E. Demidova, “Metacitta: Deep meta-learning for spatio-temporal prediction across cities and tasks,” in 27th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2023 . Springer, 2023, pp. 70–82

    2023

  31. [39]

    Multi-scale representation learning for spatial feature distributions using grid cells,

    G. Mai, K. Janowicz, B. Yan, R. Zhu, L. Cai, and N. Lao, “Multi-scale representation learning for spatial feature distributions using grid cells,” in 8th International Conference on Learning Representations, ICLR . OpenReview.net, 2020

    2020

  32. [40]

    Tile2vec: Unsupervised representation learning for spatially distributed data,

    N. Jean, S. Wang, A. Samar, G. Azzari, D. Lobell, and S. Ermon, “Tile2vec: Unsupervised representation learning for spatially distributed data,” in Proceedings of the AAAI Conference on Artificial Intelligence , vol. 33, no. 01, 2019, pp. 3967–3974

    2019

  33. [41]

    Beyond the first law of geography: Learning representations of satellite imagery by leveraging point-of-interests,

    Y . Xi, T. Li, H. Wang, Y . Li, S. Tarkoma, and P. Hui, “Beyond the first law of geography: Learning representations of satellite imagery by leveraging point-of-interests,” in Proceedings of the ACM Web Conference, WWW, 2022, pp. 3308–3316

    2022

  34. [42]

    A survey on map-matching algorithms,

    P. Chao, Y . Xu, W. Hua, and X. Zhou, “A survey on map-matching algorithms,” in 31st Australasian Database Conference, ADC , vol. 12008. Springer, 2020, pp. 121–133. 11

    2020

  35. [43]

    Learning effective road network representation with hierarchical graph neural networks,

    N. Wu, W. X. Zhao, J. Wang, and D. Pan, “Learning effective road network representation with hierarchical graph neural networks,” in 26th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD. ACM, 2020, pp. 6–14

    2020

  36. [44]

    On representation learning for road networks,

    M. Wang, W. Lee, T. Fu, and G. Yu, “On representation learning for road networks,” ACM Trans. Intell. Syst. Technol. , vol. 12, no. 1, pp. 11:1–11:27, 2021

    2021

  37. [45]

    Unsupervised path representation learning with curriculum negative sampling,

    S. B. Yang, C. Guo, J. Hu, J. Tang, and B. Yang, “Unsupervised path representation learning with curriculum negative sampling,” in 30th International Joint Conference on Artificial Intelligence, IJCAI, Z. Zhou, Ed. ijcai.org, 2021

    2021

  38. [46]

    More than routing: Joint GPS and route modeling for refine trajectory representation learning,

    Z. Ma, Z. Tu, X. Chen, Y . Zhang, D. Xia, G. Zhou, Y . Chen, Y . Zheng, and J. Gong, “More than routing: Joint GPS and route modeling for refine trajectory representation learning,” in Proceedings of the ACM Web Conference, WWW. ACM, 2024

    2024

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.