Masked autoencoder for multiagent trajectories

Research output: Journal contributionsJournal articlesResearchpeer-review

Standard

Masked autoencoder for multiagent trajectories. / Rudolph, Yannick; Brefeld, Ulf.
In: Machine Learning, Vol. 114, No. 2, 44, 02.2025.

Research output: Journal contributionsJournal articlesResearchpeer-review

Harvard

APA

Vancouver

Rudolph Y, Brefeld U. Masked autoencoder for multiagent trajectories. Machine Learning. 2025 Feb;114(2):44. doi: 10.1007/s10994-024-06647-3

Bibtex

@article{3e1d793e005841b08172ca389ba19250,
title = "Masked autoencoder for multiagent trajectories",
abstract = "Automatically labeling trajectories of multiple agents is key to behavioral analyses but usually requires a large amount of manual annotations. This also applies to the domain of team sport analyses. In this paper, we specifically show how pretraining transformer models improves the classification performance on tracking data from professional soccer. For this purpose, we propose a novel self-supervised masked autoencoder for multiagent trajectories to effectively learn from only a few labeled sequences. Our approach builds upon a factorized transformer architecture for multiagent trajectory data and employs a masking scheme on the level of individual agent trajectories. As a result, our model allows for a reconstruction of masked trajectory segments while being permutation equivariant with respect to the agent trajectories. In addition to experiments on soccer, we demonstrate the usefulness of the proposed pretraining approach on multiagent pose data from entomology. In contrast to related work, our approach is conceptually much simpler, does not require handcrafted features and naturally allows for permutation invariance in downstream tasks.",
keywords = "Business informatics, Self-supervised learning, Multiagent trajectories, Masked autoencoder, Transformer, Tracking data, Soccer, Informatics, Masked autoencoder, Multiagent trajectories, Self-supervised learning, soccer, tracking data, Transformer",
author = "Yannick Rudolph and Ulf Brefeld",
note = "Publisher Copyright: {\textcopyright} The Author(s) 2025.",
year = "2025",
month = feb,
doi = "10.1007/s10994-024-06647-3",
language = "English",
volume = "114",
journal = "Machine Learning",
issn = "0885-6125",
publisher = "Springer Netherlands",
number = "2",

}

RIS

TY - JOUR

T1 - Masked autoencoder for multiagent trajectories

AU - Rudolph, Yannick

AU - Brefeld, Ulf

N1 - Publisher Copyright: © The Author(s) 2025.

PY - 2025/2

Y1 - 2025/2

N2 - Automatically labeling trajectories of multiple agents is key to behavioral analyses but usually requires a large amount of manual annotations. This also applies to the domain of team sport analyses. In this paper, we specifically show how pretraining transformer models improves the classification performance on tracking data from professional soccer. For this purpose, we propose a novel self-supervised masked autoencoder for multiagent trajectories to effectively learn from only a few labeled sequences. Our approach builds upon a factorized transformer architecture for multiagent trajectory data and employs a masking scheme on the level of individual agent trajectories. As a result, our model allows for a reconstruction of masked trajectory segments while being permutation equivariant with respect to the agent trajectories. In addition to experiments on soccer, we demonstrate the usefulness of the proposed pretraining approach on multiagent pose data from entomology. In contrast to related work, our approach is conceptually much simpler, does not require handcrafted features and naturally allows for permutation invariance in downstream tasks.

AB - Automatically labeling trajectories of multiple agents is key to behavioral analyses but usually requires a large amount of manual annotations. This also applies to the domain of team sport analyses. In this paper, we specifically show how pretraining transformer models improves the classification performance on tracking data from professional soccer. For this purpose, we propose a novel self-supervised masked autoencoder for multiagent trajectories to effectively learn from only a few labeled sequences. Our approach builds upon a factorized transformer architecture for multiagent trajectory data and employs a masking scheme on the level of individual agent trajectories. As a result, our model allows for a reconstruction of masked trajectory segments while being permutation equivariant with respect to the agent trajectories. In addition to experiments on soccer, we demonstrate the usefulness of the proposed pretraining approach on multiagent pose data from entomology. In contrast to related work, our approach is conceptually much simpler, does not require handcrafted features and naturally allows for permutation invariance in downstream tasks.

KW - Business informatics

KW - Self-supervised learning

KW - Multiagent trajectories

KW - Masked autoencoder

KW - Transformer

KW - Tracking data

KW - Soccer

KW - Informatics

KW - Masked autoencoder

KW - Multiagent trajectories

KW - Self-supervised learning

KW - soccer

KW - tracking data

KW - Transformer

UR - http://www.scopus.com/inward/record.url?scp=85217784736&partnerID=8YFLogxK

U2 - 10.1007/s10994-024-06647-3

DO - 10.1007/s10994-024-06647-3

M3 - Journal articles

VL - 114

JO - Machine Learning

JF - Machine Learning

SN - 0885-6125

IS - 2

M1 - 44

ER -

Recently viewed

Publications

  1. Robust Control of Excavation Mobile Robot with Dynamic Triangulation Vision
  2. A Robust Decoupling Estimator to Indentify Electrical Parameters for Three-Phase Permanent Magnet Synchronous Motors
  3. Stakeholder involvement for Water Framework Directive implementation in Germany
  4. Global maps of soil temperature
  5. The Potential of AutoML for Demand Forecasting
  6. Refugee Entrepreneurship and Organizing Social Integration: Insights from Organizational Theory
  7. Using ‘mixed methods’ in sustainable consumption research
  8. Impact of Auditor and Audit Firm Rotation on Accounting and Audit Quality
  9. Playing with Information
  10. Solar Rotation Velocity Determined by Coronal Bright Points – New Data and Analysis
  11. LehrerInnen als "Reflective Practitioner"
  12. Fusion protein targeting collagen and CD133 antigen
  13. Facilitative effects of introduced Pacific oysters on native macroalgae are limited by a secondary invader, the seaweed Sargassum muticum
  14. Long-Term Abandonment of Forest Management Has a Strong Impact on Tree Morphology and Wood Volume Allocation Pattern of European Beech (Fagus Sylvatica L.)
  15. Generative Phänomenologie in feministischer Perspektive
  16. Transparency and Representation of the Public Interest in Investment Treaty Arbitration
  17. More than only skin deep: Appearance self-concept predicts most of secondary school students’ self-esteem
  18. Operaismo and the Wicked Problem of Organization
  19. Silver Work
  20. Governing Climate Change by Diffusion
  21. Warum Selbstevaluation?
  22. Historie
  23. Anamnesis and Re-Orientation
  24. Compression creep at 240°C of extruded magnesium alloys containing gadolinium
  25. (Angehende) Lehrerinnen und Lehrer als "Reflective Practitioner"
  26. Gewalt
  27. Irritierte Routinen
  28. Lamech
  29. Das Nordfenster
  30. Gewaltkarrieren von Mädchen
  31. Is there a diagrammatic impulse with Plato? ‘Quasi-diagrammatic-scenes’ in Plato’s philosophy