pith. machine review for the scientific record. sign in

arxiv: 2605.06906 · v1 · submitted 2026-05-07 · 💻 cs.LG

Recognition: no theorem link

TraXion: Rethinking Pre-training Frameworks for Mobility and Beyond

Shang-Ling Hsu , Mark Tenzer , Cyrus Shahabi , Khurram Shafique
Authors on Pith no claims yet

Pith reviewed 2026-05-11 00:55 UTC · model grok-4.3

classification 💻 cs.LG
keywords pre-trainingmobility modelingevent streamsanomaly detectionnext-visit predictiontransfer learningmulti-entity dataspatiotemporal events
0
0 comments X

The pith

TraXion is a pre-training framework built to satisfy three axioms for multi-entity spatiotemporal event streams, allowing one checkpoint per dataset to outperform task-specific models on mobility tasks and transfer directly to security and

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

Existing pre-training methods for mobility data treat trajectories like text sentences, which ignores three key structural differences: visits carry joint meaning over location time and activity, users have persistent signatures across trajectories, and visits are not independent because users share places. The paper formalizes these differences as three axioms that any pre-training framework for the broader class of multi-entity spatiotemporal event streams must meet. TraXion is designed with objectives and architecture that jointly satisfy the axioms. A single TraXion checkpoint per dataset beats task-specific baselines on every task across six public mobility datasets for anomaly detection, next-POI recommendation, next-visit prediction, and social-link prediction. The same unchanged recipe matches or exceeds prior results on enterprise authentication logs and ICU mortality prediction.

Core claim

TraXion is a pre-training framework whose objectives and architecture are jointly designed to satisfy three axioms derived from the properties of MESES data: tuple-valued events whose meaning depends on the joint distribution over location time and activity, persistent user signatures across trajectories, and non-independence across users due to co-location at shared places. A single TraXion checkpoint per dataset beats task-specific baselines on every task across six public mobility datasets covering anomaly detection, next-POI recommendation, next-visit prediction, and social-link prediction. The same recipe, applied unchanged to enterprise authentication logs and ICU mortality prediction,

What carries the argument

TraXion's objectives and architecture jointly designed to satisfy the three axioms for multi-entity spatiotemporal event streams (MESES).

If this is right

  • Mobility tasks such as anomaly detection and next-visit prediction can share one pre-trained representation instead of requiring separate models for each task.
  • Event streams from security and healthcare can be modeled with the same pre-training recipe as mobility data without any changes.
  • Performance improvements arise because the pre-training explicitly accounts for joint distributions over event attributes, persistent entity signatures, and inter-entity interactions.
  • Cross-domain transfer becomes possible once the pre-training respects the shared structural properties of MESES data rather than importing objectives from unrelated domains.

Where Pith is reading between the lines

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

  • The same axiomatic approach could extend to other MESES-like data such as financial transactions or sensor networks where entities interact through shared infrastructure.
  • If the axioms prove general, future models for sequential event data may need less domain-specific engineering because the core structure is captured at the pre-training stage.
  • Testing which of the three axioms contributes most to gains on particular tasks would clarify whether all three are required or if subsets suffice for narrower applications.

Load-bearing premise

The three structural properties of MESES data can be turned into axioms whose satisfaction by TraXion's objectives and architecture is both necessary and sufficient for the observed performance gains, with no domain-specific post-processing or hyper-parameter search required.

What would settle it

Training TraXion on a new MESES dataset from a different domain and finding that it underperforms a task-specific baseline on at least one task or requires domain-specific hyperparameter tuning to match prior results.

Figures

Figures reproduced from arXiv: 2605.06906 by Cyrus Shahabi, Khurram Shafique, Mark Tenzer, Shang-Ling Hsu.

Figure 1
Figure 1. Figure 1: Pre-training architecture of TraXion. A perturbation operator η flags a fraction of events in window Eu and corrupts their context, time, or both, yielding (E˜ u, yu). Each event becomes F=5 feature tokens—spatial (Space2Vec), two temporal (Time2Vec), the entity prototype pu, and activity—and is stacked with up to C−1 co-occurring peer events from other entities at the same substrate context x. The backbon… view at source ↗
Figure 2
Figure 2. Figure 2: UA-Berlin pre-training validation loss under three wrap settings (identical recipe, [PITH_FULL_IMAGE:figures/full_fig_p019_2.png] view at source ↗
read the original abstract

Human mobility differs from text and from generic time series in three structural ways: visits are tuple-valued events whose meaning depends on the joint distribution over location, time, and activity; users carry persistent signatures across trajectories; and visits are not independent across users, since co-location at shared places is a primary signal. Existing pre-training recipes for mobility import objectives from language modeling, treating trajectories as sentences and visits as tokens, an analogy that fails against each of the three properties above. These properties define a broader class, multi-entity spatiotemporal event streams (MESES), spanning enterprise authentication logs, electronic health records, and other event-stream domains where entities share infrastructure, schedules, or contexts. We make the properties precise as three axioms that any pre-training framework for MESES should satisfy, and introduce TraXion, whose objectives and architecture are jointly designed to meet them. A single TraXion checkpoint per dataset beats task-specific baselines on every task across six public mobility datasets covering anomaly detection, next-POI recommendation, next-visit prediction, and social-link prediction. The same recipe, applied unchanged to enterprise authentication logs and ICU mortality prediction, matches or exceeds prior work on both, showing that event streams from domains as different as mobility, security, and healthcare can be modeled under a single framework.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 0 minor

Summary. The paper claims that mobility trajectories and similar multi-entity spatiotemporal event streams (MESES) exhibit three structural properties—tuple-valued events whose semantics depend on joint location-time-activity distributions, persistent user signatures across trajectories, and non-independence of visits across users due to co-location—that distinguish them from text or generic time series. Existing language-modeling-style pre-training fails to respect these properties. The authors formalize the properties as three axioms that any MESES pre-training framework should satisfy, introduce TraXion whose objectives and architecture are jointly designed to meet the axioms, and report that a single TraXion checkpoint per dataset outperforms task-specific baselines on anomaly detection, next-POI recommendation, next-visit prediction, and social-link prediction across six public mobility datasets. The identical recipe, applied unchanged, matches or exceeds prior work on enterprise authentication logs and ICU mortality prediction.

Significance. If the empirical results and the necessity of the axioms are substantiated, the work would be significant: it offers a unified pre-training recipe for event-stream domains that share infrastructure or context (mobility, security, healthcare) and avoids the common practice of importing language-modeling objectives that ignore the data's joint structure and cross-entity dependencies.

major comments (3)
  1. [Abstract] Abstract and experimental section: the central claim that a single TraXion checkpoint beats every task-specific baseline on all tasks requires a reproducible experimental protocol, baseline implementations, statistical significance tests, and ablation results; none of these are described or referenced in the provided text, rendering the performance claims unverifiable.
  2. [Abstract] The argument that the three MESES axioms drive the observed gains (rather than other unablated factors such as encoder choice or training schedule) is load-bearing for the paper's contribution, yet no ablation is reported that removes or relaxes one axiom while keeping the rest of the architecture and recipe fixed and then measures the resulting performance drop on the reported tasks.
  3. [Abstract] Abstract: the cross-domain transfer result (authentication logs and ICU mortality) is presented as using the 'same recipe, applied unchanged,' but no details are given on whether hyper-parameter search, domain-specific post-processing, or re-tuning occurred; this information is necessary to evaluate the claim of zero-shot transfer.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. The comments highlight important gaps in experimental documentation and validation that we will address in the revision. Below we respond point-by-point to the three major comments.

read point-by-point responses

  1. Referee: [Abstract] Abstract and experimental section: the central claim that a single TraXion checkpoint beats every task-specific baseline on all tasks requires a reproducible experimental protocol, baseline implementations, statistical significance tests, and ablation results; none of these are described or referenced in the provided text, rendering the performance claims unverifiable.

    Authors: We agree that the current manuscript version lacks sufficient detail for independent reproduction of the results. In the revised manuscript we will add a dedicated 'Experimental Setup' subsection that fully specifies the training protocol (including data splits, batching, optimization hyperparameters, and early-stopping criteria), provides references or links to the exact baseline implementations used, reports statistical significance (paired t-tests or Wilcoxon signed-rank tests with p-values across five random seeds), and includes the requested ablation results. These additions will make the performance claims verifiable. revision: yes

  2. Referee: [Abstract] The argument that the three MESES axioms drive the observed gains (rather than other unablated factors such as encoder choice or training schedule) is load-bearing for the paper's contribution, yet no ablation is reported that removes or relaxes one axiom while keeping the rest of the architecture and recipe fixed and then measures the resulting performance drop on the reported tasks.

    Authors: We acknowledge that the manuscript does not yet contain ablations that isolate the contribution of each axiom. In the revision we will introduce three controlled variants of TraXion, each violating exactly one axiom while preserving the encoder, training schedule, and all other objectives: (1) an independent-event variant that factorizes the joint location-time-activity distribution, (2) a user-agnostic variant that drops persistent signature modeling, and (3) a user-independent variant that ignores co-location signals. We will report the resulting performance drops on all four mobility tasks, thereby directly linking the axioms to the observed gains. revision: yes

  3. Referee: [Abstract] Abstract: the cross-domain transfer result (authentication logs and ICU mortality) is presented as using the 'same recipe, applied unchanged,' but no details are given on whether hyper-parameter search, domain-specific post-processing, or re-tuning occurred; this information is necessary to evaluate the claim of zero-shot transfer.

    Authors: We will clarify this point explicitly. The transfer experiments used the identical pre-training objectives, architecture, optimizer settings, and hyperparameter values as the mobility runs; the only adaptation was the minimal input tokenization required to map the new event schemas into the same vocabulary format. No hyper-parameter search, domain-specific fine-tuning, or post-processing was performed on the target domains. A new paragraph in the experimental section will document this procedure and confirm the zero-shot nature of the transfer. revision: yes

Circularity Check

0 steps flagged

No circularity: axioms are motivated externally and performance is empirical.

full rationale

The paper extracts three structural properties from MESES data descriptions, formalizes them as axioms, designs objectives and architecture to satisfy those axioms, and reports experimental results on held-out tasks and cross-domain datasets. No equation equates a derived quantity to a fitted parameter defined by the same metric, no self-citation supplies a load-bearing uniqueness theorem, and no ansatz is smuggled via prior work. The performance edge is presented as an observed outcome of the design rather than a mathematical identity with the inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 2 invented entities

Only the abstract is available. The central claim rests on three axioms derived from mobility properties and on the assertion that TraXion satisfies them. No numerical free parameters, additional axioms, or invented physical entities are extractable from the given text.

axioms (3)
  • domain assumption Visits are tuple-valued events whose meaning depends on the joint distribution over location, time, and activity.
    Stated in abstract as one of the three structural properties that define MESES and that existing language-model objectives fail to respect.
  • domain assumption Users carry persistent signatures across trajectories.
    Stated in abstract as second structural property that pre-training must capture.
  • domain assumption Visits are not independent across users because co-location at shared places is a primary signal.
    Stated in abstract as third structural property that pre-training must respect.
invented entities (2)
  • MESES (multi-entity spatiotemporal event streams) no independent evidence
    purpose: Broader data class that includes mobility, authentication logs, and electronic health records.
    Newly introduced term in the abstract to unify the domains.
  • TraXion no independent evidence
    purpose: Pre-training framework whose objectives and architecture satisfy the three MESES axioms.
    Newly introduced model name and design in the abstract.

pith-pipeline@v0.9.0 · 5535 in / 1723 out tokens · 54818 ms · 2026-05-11T00:55:14.064945+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

81 extracted references · 6 canonical work pages · 1 internal anchor

  1. [1]

    Urban Anomalies: A simulated human mobility dataset with injected anomalies

    Hossein Amiri, Ruochen Kong, and Andreas Züfle. Urban Anomalies: A simulated human mobility dataset with injected anomalies. InProceedings of the 1st ACM SIGSPATIAL International Workshop on Geospatial Anomaly Detection, pages 1–11, 2024

    2024

  2. [2]

    Claude Code.https://www.anthropic.com/claude-code, 2025

    Anthropic. Claude Code.https://www.anthropic.com/claude-code, 2025

    2025

  3. [3]

    ICAD: A self-supervised autoregressive approach for multi-context anomaly detection in human mobility data

    Bita Azarijoo, Maria Despoina Siampou, John Krumm, and Cyrus Shahabi. ICAD: A self-supervised autoregressive approach for multi-context anomaly detection in human mobility data. InProceedings of 9 the 33rd ACM International Conference on Advances in Geographic Information Systems, pages 595–606, 2025

    2025

  4. [4]

    Contrastive trajectory similarity learning with dual-feature attention

    Yanchuan Chang, Jianzhong Qi, Yuxuan Liang, and Egemen Tanin. Contrastive trajectory similarity learning with dual-feature attention. In2023 IEEE 39th International Conference on Data Engineering (ICDE), pages 2933–2945. IEEE, 2023

    2023

  5. [5]

    Recurrent neural networks for multivariate time series with missing values.Scientific Reports, 8(1):6085, 2018

    Zhengping Che, Sanjay Purushotham, Kyunghyun Cho, David Sontag, and Yan Liu. Recurrent neural networks for multivariate time series with missing values.Scientific Reports, 8(1):6085, 2018

    2018

  6. [6]

    Mutual distillation learn- ing network for trajectory-user linking

    Wei Chen, Shuzhe Li, Chao Huang, Yanwei Yu, Yongguo Jiang, and Junyu Dong. Mutual distillation learn- ing network for trajectory-user linking. InProceedings of the Thirty-First International Joint Conference on Artificial Intelligence. International Joint Conferences on Artificial Intelligence Organization., 2022

    2022

  7. [7]

    Trajectory-user linking via hierarchical spatio-temporal attention networks.ACM Transactions on Knowledge Discovery from Data, 18(4):1–22, 2024

    Wei Chen, Chao Huang, Yanwei Yu, Yongguo Jiang, and Junyu Dong. Trajectory-user linking via hierarchical spatio-temporal attention networks.ACM Transactions on Knowledge Discovery from Data, 18(4):1–22, 2024

    2024

  8. [8]

    Friendship and mobility: user movement in location- based social networks

    Eunjoon Cho, Seth A Myers, and Jure Leskovec. Friendship and mobility: user movement in location- based social networks. InProceedings of the 17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 1082–1090, 2011

    2011

  9. [9]

    One model, many cities: A transferable social relationship inference framework for human mobility data

    Chen Chu, Cyrus Shahabi, Emmanuel Tung, and Khurram Shafique. One model, many cities: A transferable social relationship inference framework for human mobility data. InProceedings of the 33rd ACM International Conference on Advances in Geographic Information Systems, pages 66–76, 2025

    2025

  10. [10]

    Le, and Christopher D

    Kevin Clark, Minh-Thang Luong, Quoc V . Le, and Christopher D. Manning. ELECTRA: Pre-training text encoders as discriminators rather than generators. InInternational Conference on Learning Representations, 2020

    2020

  11. [11]

    BERT: Pre-training of deep bidirectional transformers for language understanding

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep bidirectional transformers for language understanding. InProceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 4171–4186, 2019

    2019

  12. [12]

    SimMTM: A simple pre-training framework for masked time-series modeling.Advances in Neural Information Processing Systems, 36:29996–30025, 2023

    Jiaxiang Dong, Haixu Wu, Haoran Zhang, Li Zhang, Jianmin Wang, and Mingsheng Long. SimMTM: A simple pre-training framework for masked time-series modeling.Advances in Neural Information Processing Systems, 36:29996–30025, 2023

    2023

  13. [13]

    DeepLog: Anomaly detection and diagnosis from system logs through deep learning

    Min Du, Feifei Li, Guineng Zheng, and Vivek Srikumar. DeepLog: Anomaly detection and diagnosis from system logs through deep learning. InProceedings of the 2017 ACM SIGSAC Conference on Computer and Communications Security, pages 1285–1298, 2017

    2017

  14. [14]

    Dirichlet-Hawkes processes with applications to clustering continuous-time document streams

    Nan Du, Mehrdad Farajtabar, Amr Ahmed, Alexander J Smola, and Le Song. Dirichlet-Hawkes processes with applications to clustering continuous-time document streams. InProceedings of the 21st ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 219–228, 2015

    2015

  15. [15]

    Recurrent marked temporal point processes: Embedding event history to vector

    Nan Du, Hanjun Dai, Rakshit Trivedi, Utkarsh Upadhyay, Manuel Gomez-Rodriguez, and Le Song. Recurrent marked temporal point processes: Embedding event history to vector. InProceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 1555–1564, 2016

    2016

  16. [16]

    Back to Bayesics: Uncovering human mobility distributions and anomalies with an integrated statistical and neural framework

    Minxuan Duan, Yinlong Qian, Lingyi Zhao, Zihao Zhou, Zeeshan Rasheed, Rose Yu, and Khurram Shafique. Back to Bayesics: Uncovering human mobility distributions and anomalies with an integrated statistical and neural framework. InProceedings of the 1st ACM SIGSPATIAL International Workshop on Geospatial Anomaly Detection, pages 56–67, 2024

    2024

  17. [17]

    AgentMove: A large language model based agentic framework for zero-shot next location prediction

    Jie Feng, Yuwei Du, Jie Zhao, and Yong Li. AgentMove: A large language model based agentic framework for zero-shot next location prediction. InProceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pages 1322–1338, 2025

    2025

  18. [18]

    Improving event representa- tion via simultaneous weakly supervised contrastive learning and clustering

    Jun Gao, Wei Wang, Changlong Yu, Huan Zhao, Wilfred Ng, and Ruifeng Xu. Improving event representa- tion via simultaneous weakly supervised contrastive learning and clustering. InProceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2022. 10

    2022

  19. [19]

    Letian Gong, Yan Lin, Xinyue Zhang, Yiwen Lu, Xuedi Han, Yichen Liu, Shengnan Guo, Youfang Lin, and Huaiyu Wan. Mobility-LLM: Learning visiting intentions and travel preference from human mobility data with large language models.Advances in Neural Information Processing Systems, 37:36185–36217, 2024

    2024

  20. [20]

    TabR: Tabular deep learning meets nearest neighbors

    Yury Gorishniy, Ivan Rubachev, Nikolay Kartashev, Daniil Shlenskii, Akim Kotelnikov, and Artem Babenko. TabR: Tabular deep learning meets nearest neighbors. InThe Twelfth International Conference on Learning Representations, 2024

    2024

  21. [21]

    LogBERT: Log anomaly detection via BERT

    Haixuan Guo, Shuhan Yuan, and Xintao Wu. LogBERT: Log anomaly detection via BERT. In2021 International Joint Conference on Neural Networks (IJCNN), pages 1–8. IEEE, 2021

    2021

  22. [22]

    WaveGNN: Integrating graph neural networks and transformers for decay-aware classification of irregular clinical time-series

    Arash Hajisafi, Maria Despoina Siampou, Bita Azarijoo, Zhen Xiong, and Cyrus Shahabi. WaveGNN: Integrating graph neural networks and transformers for decay-aware classification of irregular clinical time-series. In2025 IEEE International Conference on Big Data (BigData), pages 1934–1943, 2025. doi: 10.1109/BigData66926.2025.11401906

    work page doi:10.1109/bigdata66926.2025.11401906 1934

  23. [23]

    LogGPT: Log anomaly detection via GPT

    Xiao Han, Shuhan Yuan, and Mohamed Trabelsi. LogGPT: Log anomaly detection via GPT. In2023 IEEE International Conference on Big Data (BigData), pages 1117–1122, 2023

    2023

  24. [24]

    MobilityGPT: Enhanced human mobility modeling with a GPT model.IEEE Transactions on Intelligent Transportation Systems, 27(1):1681–1694, 2025

    Ammar Haydari, Dongjie Chen, Zhengfeng Lai, Michael Zhang, and Chen-Nee Chuah. MobilityGPT: Enhanced human mobility modeling with a GPT model.IEEE Transactions on Intelligent Transportation Systems, 27(1):1681–1694, 2025

    2025

  25. [25]

    Masked autoencoders are scalable vision learners

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, and Ross Girshick. Masked autoencoders are scalable vision learners. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 16000–16009, 2022

    2022

  26. [26]

    TrajGPT: Controlled synthetic trajectory generation using a multitask transformer-based spatiotemporal model

    Shang-Ling Hsu, Emmanuel Tung, John Krumm, Cyrus Shahabi, and Khurram Shafique. TrajGPT: Controlled synthetic trajectory generation using a multitask transformer-based spatiotemporal model. In Proceedings of the 32nd ACM International Conference on Advances in Geographic Information Systems, pages 362–371, 2024

    2024

  27. [27]

    eICU Collaborative Research Database Demo.PhysioNet, May 2021

    Alistair Johnson, Tom Pollard, Omar Badawi, and Jesse Raffa. eICU Collaborative Research Database Demo.PhysioNet, May 2021. doi: 10.13026/4mxk-na84. URL https://doi.org/10.13026/ 4mxk-na84. Version 2.0.1

    work page doi:10.13026/4mxk-na84 2021

  28. [28]

    Self-attentive sequential recommendation

    Wang-Cheng Kang and Julian McAuley. Self-attentive sequential recommendation. In2018 IEEE International Conference on Data Mining (ICDM), pages 197–206, 2018. doi: 10.1109/ICDM.2018.00035

    work page doi:10.1109/icdm.2018.00035 2018

  29. [29]

    autoresearch.https://github.com/karpathy/autoresearch, 2025

    Andrej Karpathy. autoresearch.https://github.com/karpathy/autoresearch, 2025

    2025

  30. [30]

    Time2Vec: Learning a Vector Representation of Time

    Seyed Mehran Kazemi, Rishab Goel, Sepehr Eghbali, Janahan Ramanan, Jaspreet Sahota, Sanjay Thakur, Stella Wu, Cathal Smyth, Pascal Poupart, and Marcus Brubaker. Time2Vec: Learning a vector representa- tion of time.arXiv preprint arXiv:1907.05321, 2019

    work page Pith review arXiv 1907

  31. [31]

    Comprehensive, multi-source cyber-security events data set

    Alexander D Kent. Comprehensive, multi-source cyber-security events data set. Technical report, Los Alamos National Lab.(LANL), Los Alamos, NM (United States), 2015

    2015

  32. [32]

    Similarity of neural network representations revisited

    Simon Kornblith, Mohammad Norouzi, Honglak Lee, and Geoffrey Hinton. Similarity of neural network representations revisited. InInternational Conference on Machine Learning, 2019

    2019

  33. [33]

    Junnan Li, Pan Zhou, Caiming Xiong, and Steven C.H. Hoi. Prototypical contrastive learning of unsuper- vised representations. InInternational Conference on Learning Representations, 2021

    2021

  34. [34]

    TrajFlow: Nation-wide pseudo GPS trajectory generation with flow matching models

    Peiran Li, Jiawei Wang, Haoran Zhang, Xiaodan Shi, Noboru Koshizuka, Chihiro Shimizu, and Renhe Jiang. TrajFlow: Nation-wide pseudo GPS trajectory generation with flow matching models. InThe Fourteenth International Conference on Learning Representations, 2026

    2026

  35. [35]

    BEHRT: transformer for electronic health records.Scientific Reports, 10(1):7155, 2020

    Yikuan Li, Shishir Rao, José Roberto Ayala Solares, Abdelaali Hassaine, Rema Ramakrishnan, Dexter Canoy, Yajie Zhu, Kazem Rahimi, and Gholamreza Salimi-Khorshidi. BEHRT: transformer for electronic health records.Scientific Reports, 10(1):7155, 2020

    2020

  36. [36]

    Heterogeneous hyperbolic hypergraph neural network for friend recommendation in location-based social networks.ACM Transactions on Knowledge Discovery from Data, 19(3):1–29, 2025

    Yongkang Li, Zipei Fan, and Xuan Song. Heterogeneous hyperbolic hypergraph neural network for friend recommendation in location-based social networks.ACM Transactions on Knowledge Discovery from Data, 19(3):1–29, 2025

    2025

  37. [37]

    Temporal fusion transformers for interpretable multi-horizon time series forecasting.International Journal of Forecasting, 37(4):1748–1764, 2021

    Bryan Lim, Sercan Ö Arık, Nicolas Loeff, and Tomas Pfister. Temporal fusion transformers for interpretable multi-horizon time series forecasting.International Journal of Forecasting, 37(4):1748–1764, 2021. 11

    2021

  38. [38]

    Pre-training context and time aware location embeddings from spatial-temporal trajectories for user next location prediction

    Yan Lin, Huaiyu Wan, Shengnan Guo, and Youfang Lin. Pre-training context and time aware location embeddings from spatial-temporal trajectories for user next location prediction. InProceedings of the AAAI Conference on Artificial Intelligence, volume 35, pages 4241–4248, 2021

    2021

  39. [39]

    Discovering latent network structure in point process data

    Scott Linderman and Ryan Adams. Discovering latent network structure in point process data. In International Conference on Machine Learning, pages 1413–1421. PMLR, 2014

    2014

  40. [40]

    iTrans- former: Inverted transformers are effective for time series forecasting

    Yong Liu, Tengge Hu, Haoran Zhang, Haixu Wu, Shiyu Wang, Lintao Ma, and Mingsheng Long. iTrans- former: Inverted transformers are effective for time series forecasting. InThe Twelfth International Conference on Learning Representations, 2024

    2024

  41. [41]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. InInternational Conference on Learning Representations (ICLR), 2019

    2019

  42. [42]

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

    Gengchen Mai, Krzysztof Janowicz, Bo Yan, Rui Zhu, Ling Cai, and Ni Lao. Multi-scale representation learning for spatial feature distributions using grid cells. InInternational Conference on Learning Representations (ICLR), 2020

    2020

  43. [43]

    UMAP: Uniform manifold approxi- mation and projection.Journal of Open Source Software, 3(29), 2018

    Leland McInnes, John Healy, Nathaniel Saul, and Lukas Großberger. UMAP: Uniform manifold approxi- mation and projection.Journal of Open Source Software, 3(29), 2018

    2018

  44. [44]

    The neural Hawkes process: A neurally self-modulating multivariate point process.Advances in Neural Information Processing Systems, 30, 2017

    Hongyuan Mei and Jason M Eisner. The neural Hawkes process: A neurally self-modulating multivariate point process.Advances in Neural Information Processing Systems, 30, 2017

    2017

  45. [45]

    Self-supervised log parsing

    Sasho Nedelkoski, Jasmin Bogatinovski, Alexander Acker, Jorge Cardoso, and Odej Kao. Self-supervised log parsing. InJoint European Conference on Machine Learning and Knowledge Discovery in Databases (ECML-PKDD), 2020

    2020

  46. [46]

    A time series is worth 64 words: Long-term forecasting with transformers

    Yuqi Nie, Nam H Nguyen, Phanwadee Sinthong, and Jayant Kalagnanam. A time series is worth 64 words: Long-term forecasting with transformers. InThe Eleventh International Conference on Learning Representations, 2023

    2023

  47. [47]

    CEHR-BERT: Incorporating temporal information from structured ehr data to improve prediction tasks

    Chao Pang, Xinzhuo Jiang, Krishna S Kalluri, Matthew Spotnitz, RuiJun Chen, Adler Perotte, and Karthik Natarajan. CEHR-BERT: Incorporating temporal information from structured ehr data to improve prediction tasks. InMachine Learning for Health, pages 239–260. PMLR, 2021

    2021

  48. [48]

    EBM: an entropy-based model to infer social strength from spatiotemporal data

    Huy Pham, Cyrus Shahabi, and Yan Liu. EBM: an entropy-based model to infer social strength from spatiotemporal data. InProceedings of the 2013 ACM SIGMOD International Conference on Management of Data, pages 265–276, 2013

    2013

  49. [49]

    Med-BERT: pretrained contextualized embeddings on large-scale structured electronic health records for disease prediction.NPJ Digital Medicine, 4(1):86, 2021

    Laila Rasmy, Yang Xiang, Ziqian Xie, Cui Tao, and Degui Zhi. Med-BERT: pretrained contextualized embeddings on large-scale structured electronic health records for disease prediction.NPJ Digital Medicine, 4(1):86, 2021

    2021

  50. [50]

    Samir, Jaroslaw Was, Quanzheng Li, David W

    Pawel Renc, Yugang Jia, Anthony E. Samir, Jaroslaw Was, Quanzheng Li, David W. Bates, and Arkadiusz Sitek. Zero shot health trajectory prediction using transformer.NPJ Digital Medicine, 7(1):256, 2024

    2024

  51. [51]

    Neural temporal point processes: A review

    Oleksandr Shchur, Ali Caner Türkmen, Tim Januschowski, and Stephan Günnemann. Neural temporal point processes: A review. InProceedings of the Thirtieth International Joint Conference on Artificial Intelligence, pages 4585–4593. International Joint Conferences on Artificial Intelligence Organization, 2021

    2021

  52. [52]

    Multi-time attention networks for irregularly sampled time series

    Satya Narayan Shukla and Benjamin Marlin. Multi-time attention networks for irregularly sampled time series. InInternational Conference on Learning Representations, 2021

    2021

  53. [53]

    Mobility-Embedded POIs: Learning what a place is and how it is used from human movement

    Maria Despoina Siampou, Shushman Choudhury, Shang-Ling Hsu, Neha Arora, and Cyrus Shahabi. Mobility-Embedded POIs: Learning what a place is and how it is used from human movement. In Forty-third International Conference on Machine Learning, 2026. To appear

    2026

  54. [54]

    NUMOSIM: A synthetic mobility dataset with anomaly detection benchmarks

    Chris Stanford, Suman Adari, Xishun Liao, Yueshuai He, Qinhua Jiang, Chenchen Kuai, Jiaqi Ma, Emmanuel Tung, Yinlong Qian, Lingyi Zhao, et al. NUMOSIM: A synthetic mobility dataset with anomaly detection benchmarks. InProceedings of the 1st ACM SIGSPATIAL International Workshop on Geospatial Anomaly Detection, pages 68–78, 2024

    2024

  55. [55]

    BERT4Rec: Sequential recommendation with bidirectional encoder representations from transformer

    Fei Sun, Jun Liu, Jian Wu, Changhua Pei, Xiao Lin, Wenwu Ou, and Peng Jiang. BERT4Rec: Sequential recommendation with bidirectional encoder representations from transformer. InProceedings of the 28th ACM International Conference on Information and Knowledge Management, pages 1441–1450, 2019. 12

    2019

  56. [56]

    Self-supervised transformer for sparse and irregularly sampled multivariate clinical time-series.ACM Transactions on Knowledge Discovery from Data (TKDD), 16(6): 1–17, 2022

    Sindhu Tipirneni and Chandan K Reddy. Self-supervised transformer for sparse and irregularly sampled multivariate clinical time-series.ACM Transactions on Knowledge Discovery from Data (TKDD), 16(6): 1–17, 2022

    2022

  57. [57]

    Deep learning for unsupervised insider threat detection in structured cybersecurity data streams

    Aaron Tuor, Samuel Kaplan, Brian Hutchinson, Nicole Nichols, and Sean Robinson. Deep learning for unsupervised insider threat detection in structured cybersecurity data streams. InAAAI Workshops, pages 224–231, 2017

    2017

  58. [58]

    Representation Learning with Contrastive Predictive Coding

    Aaron van den Oord, Yazhe Li, and Oriol Vinyals. Representation learning with contrastive predictive coding.arXiv preprint arXiv:1807.03748, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  59. [59]

    Pre-training time-aware location embeddings from spatial-temporal trajectories.IEEE Transactions on Knowledge and Data Engineering, 34(11):5510–5523, 2021

    Huaiyu Wan, Yan Lin, Shengnan Guo, and Youfang Lin. Pre-training time-aware location embeddings from spatial-temporal trajectories.IEEE Transactions on Knowledge and Data Engineering, 34(11):5510–5523, 2021

    2021

  60. [60]

    CoBAD: Modeling collective behaviors for human mobility anomaly detection

    Haomin Wen, Shurui Cao, and Leman Akoglu. CoBAD: Modeling collective behaviors for human mobility anomaly detection. InProceedings of the 33rd ACM International Conference on Advances in Geographic Information Systems, pages 197–209, 2025

    2025

  61. [61]

    Uncertainty- aware spatio-temporal human mobility modeling and anomaly detection

    Haomin Wen, Shurui Cao, Zeeshan Rasheed, Khurram Hassan Shafique, and Leman Akoglu. Uncertainty- aware spatio-temporal human mobility modeling and anomaly detection. InProceedings of the 33rd ACM International Conference on Advances in Geographic Information Systems, pages 328–331, 2025

    2025

  62. [62]

    Contrastive learning for sequential recommendation

    Xu Xie, Fei Sun, Zhaoyang Liu, Shiwen Wu, Jinyang Gao, Jiandong Zhang, Bolin Ding, and Bin Cui. Contrastive learning for sequential recommendation. In2022 IEEE 38th International Conference on Data Engineering (ICDE), pages 1259–1273. IEEE, 2022

    2022

  63. [63]

    COAST: Contrastive learning with augmented spatio-temporal encoding for next POI recommendation

    Bada Xin, Xin Wan, Zhuojun Jiang, Faqiang Liu, Su Chen, Rong Yang, and Qingyun Liu. COAST: Contrastive learning with augmented spatio-temporal encoding for next POI recommendation. InICASSP 2025–2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 1–5. IEEE, 2025

    2025

  64. [64]

    Taming the long tail in human mobility prediction.Advances in Neural Information Processing Systems, 37:54748–54771, 2024

    Xiaohang Xu, Renhe Jiang, Chuang Yang, Zipei Fan, and Kaoru Sezaki. Taming the long tail in human mobility prediction.Advances in Neural Information Processing Systems, 37:54748–54771, 2024

    2024

  65. [65]

    Where and when: predict next POI and its explicit timestamp in sequential recommendation

    Yuanbo Xu, Hongxu Shen, Yiheng Jiang, and En Wang. Where and when: predict next POI and its explicit timestamp in sequential recommendation. InProceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, pages 3507–3515, 2025

    2025

  66. [66]

    MobTCast: Leveraging auxiliary trajectory forecast- ing for human mobility prediction.Advances in Neural Information Processing Systems, 34:30380–30391, 2021

    Hao Xue, Flora Salim, Yongli Ren, and Nuria Oliver. MobTCast: Leveraging auxiliary trajectory forecast- ing for human mobility prediction.Advances in Neural Information Processing Systems, 34:30380–30391, 2021

    2021

  67. [67]

    Transformer embeddings of irregularly spaced events and their participants

    Chenghao Yang, Hongyuan Mei, and Jason Eisner. Transformer embeddings of irregularly spaced events and their participants. InInternational Conference on Learning Representations, 2022

    2022

  68. [68]

    Revisiting user mobility and social relationships in LBSNs: a hypergraph embedding approach

    Dingqi Yang, Bingqing Qu, Jie Yang, and Philippe Cudre-Mauroux. Revisiting user mobility and social relationships in LBSNs: a hypergraph embedding approach. InThe World Wide Web Conference, pages 2147–2157, 2019

    2019

  69. [69]

    GETNext: trajectory flow map enhanced transformer for next POI recommendation

    Song Yang, Jiamou Liu, and Kaiqi Zhao. GETNext: trajectory flow map enhanced transformer for next POI recommendation. InProceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 1144–1153, 2022

    2022

  70. [70]

    UniST: A prompt-empowered universal model for urban spatio-temporal prediction

    Yuan Yuan, Jingtao Ding, Jie Feng, Depeng Jin, and Yong Li. UniST: A prompt-empowered universal model for urban spatio-temporal prediction. InProceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 4095–4106, 2024

    2024

  71. [71]

    TS2Vec: Towards universal representation of time series

    Zhihan Yue, Yujing Wang, Juanyong Duan, Tianmeng Yang, Congrui Huang, Yunhai Tong, and Bixiong Xu. TS2Vec: Towards universal representation of time series. InProceedings of the AAAI Conference on Artificial Intelligence, volume 36, pages 8980–8987, 2022

    2022

  72. [72]

    A transformer-based framework for multivariate time series representation learning

    George Zerveas, Srideepika Jayaraman, Dhaval Patel, Anuradha Bhamidipaty, and Carsten Eickhoff. A transformer-based framework for multivariate time series representation learning. InProceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining, pages 2114–2124, 2021

    2021

  73. [73]

    Scalable trajectory-user linking with dual-stream representation networks

    Hao Zhang, Wei Chen, Xingyu Zhao, Jianpeng Qi, Guiyuan Jiang, and Yanwei Yu. Scalable trajectory-user linking with dual-stream representation networks. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 13224–13232, 2025. 13

    2025

  74. [74]

    Graph-guided network for irregularly sampled multivariate time series

    Xiang Zhang, Marko Zeman, Theodoros Tsiligkaridis, and Marinka Zitnik. Graph-guided network for irregularly sampled multivariate time series. InInternational Conference on Learning Representations, 2022

    2022

  75. [75]

    Crossformer: Transformer utilizing cross-dimension dependency for multivariate time series forecasting

    Yunhao Zhang and Junchi Yan. Crossformer: Transformer utilizing cross-dimension dependency for multivariate time series forecasting. InThe Eleventh International Conference on Learning Representations, 2023

    2023

  76. [76]

    Trajectory-user linking via variational autoencoder

    Fan Zhou, Qiang Gao, Goce Trajcevski, Kunpeng Zhang, Ting Zhong, and Fengli Zhang. Trajectory-user linking via variational autoencoder. InProceedings of the 27th International Joint Conference on Artificial Intelligence, pages 3212–3218, 2018

    2018

  77. [77]

    Learning triggering kernels for multi-dimensional Hawkes processes

    Ke Zhou, Hongyuan Zha, and Le Song. Learning triggering kernels for multi-dimensional Hawkes processes. InInternational Conference on Machine Learning, pages 1301–1309. PMLR, 2013

    2013

  78. [78]

    S3-Rec: Self-supervised learning for sequential recommendation with mutual information maximization

    Kun Zhou, Hui Wang, Wayne Xin Zhao, Yutao Zhu, Sirui Wang, Fuzheng Zhang, Zhongyuan Wang, and Ji-Rong Wen. S3-Rec: Self-supervised learning for sequential recommendation with mutual information maximization. InProceedings of the 29th ACM International Conference on Information & Knowledge Management, pages 1893–1902, 2020

    1902

  79. [79]

    One fits all: Power general time series analysis by pretrained LM.Advances in Neural Information Processing Systems, 36:43322–43355, 2023

    Tian Zhou, Peisong Niu, Liang Sun, Rong Jin, et al. One fits all: Power general time series analysis by pretrained LM.Advances in Neural Information Processing Systems, 36:43322–43355, 2023

    2023

  80. [80]

    UniTraj: Learning a universal trajectory foundation model from billion-scale worldwide traces

    Yuanshao Zhu, James Jianqiao Yu, Xiangyu Zhao, Xun Zhou, Liang Han, Xuetao Wei, and Yuxuan Liang. UniTraj: Learning a universal trajectory foundation model from billion-scale worldwide traces. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025

    2025

Showing first 80 references.