MetaCitta: Deep Meta-Learning for Spatio-Temporal Prediction Across Cities and Tasks

Research output: Chapter in book/report/conference proceedingConference contributionResearchpeer review

Authors

Research Organisations

External Research Organisations

  • Volkswagen AG
  • University of Bonn
  • Lamarr Institute for Machine Learning and Artificial Intelligence
View graph of relations

Details

Original languageEnglish
Title of host publicationAdvances in Knowledge Discovery and Data Mining
Subtitle of host publication27th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2023, Osaka, Japan, May 25–28, 2023, Proceedings, Part IV
EditorsHisashi Kashima, Tsuyoshi Ide, Wen-Chih Peng
PublisherSpringer Science and Business Media Deutschland GmbH
Pages70-82
Number of pages13
ISBN (electronic)978-3-031-33383-5
ISBN (print)9783031333828
Publication statusPublished - 26 May 2023
Event27th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2023 - Osaka, Japan
Duration: 25 May 202328 May 2023

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume13938 LNCS
ISSN (Print)0302-9743
ISSN (electronic)1611-3349

Abstract

Accurate spatio-temporal prediction is essential for capturing city dynamics and planning mobility services. State-of-the-art deep spatio-temporal predictive models depend on rich and representative training data for target regions and tasks. However, the availability of such data is typically limited. Furthermore, existing predictive models fail to utilize cross-correlations across tasks and cities. In this paper, we propose MetaCitta, a novel deep meta-learning approach that addresses the critical challenges of data scarcity and model generalization. MetaCitta adopts the data from different cities and tasks in a generalizable spatio-temporal deep neural network. We propose a novel meta-learning algorithm that minimizes the discrepancy between spatio-temporal representations across tasks and cities. Our experiments with real-world data demonstrate that the proposed MetaCitta approach outperforms state-of-the-art prediction methods for zero-shot learning and pre-training plus fine-tuning. Furthermore, MetaCitta is computationally more efficient than the existing meta-learning approaches.

Keywords

    Meta-Learning, Pre-training, Spatio-Temporal Prediction

ASJC Scopus subject areas

Cite this

MetaCitta: Deep Meta-Learning for Spatio-Temporal Prediction Across Cities and Tasks. / Sao, Ashutosh; Gottschalk, Simon; Tempelmeier, Nicolas et al.
Advances in Knowledge Discovery and Data Mining: 27th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2023, Osaka, Japan, May 25–28, 2023, Proceedings, Part IV. ed. / Hisashi Kashima; Tsuyoshi Ide; Wen-Chih Peng. Springer Science and Business Media Deutschland GmbH, 2023. p. 70-82 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13938 LNCS).

Research output: Chapter in book/report/conference proceedingConference contributionResearchpeer review

Sao, A, Gottschalk, S, Tempelmeier, N & Demidova, E 2023, MetaCitta: Deep Meta-Learning for Spatio-Temporal Prediction Across Cities and Tasks. in H Kashima, T Ide & W-C Peng (eds), Advances in Knowledge Discovery and Data Mining: 27th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2023, Osaka, Japan, May 25–28, 2023, Proceedings, Part IV. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 13938 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 70-82, 27th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2023, Osaka, Japan, 25 May 2023. https://doi.org/10.1007/978-3-031-33383-5_6
Sao, A., Gottschalk, S., Tempelmeier, N., & Demidova, E. (2023). MetaCitta: Deep Meta-Learning for Spatio-Temporal Prediction Across Cities and Tasks. In H. Kashima, T. Ide, & W.-C. Peng (Eds.), Advances in Knowledge Discovery and Data Mining: 27th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2023, Osaka, Japan, May 25–28, 2023, Proceedings, Part IV (pp. 70-82). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13938 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-031-33383-5_6
Sao A, Gottschalk S, Tempelmeier N, Demidova E. MetaCitta: Deep Meta-Learning for Spatio-Temporal Prediction Across Cities and Tasks. In Kashima H, Ide T, Peng WC, editors, Advances in Knowledge Discovery and Data Mining: 27th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2023, Osaka, Japan, May 25–28, 2023, Proceedings, Part IV. Springer Science and Business Media Deutschland GmbH. 2023. p. 70-82. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-031-33383-5_6
Sao, Ashutosh ; Gottschalk, Simon ; Tempelmeier, Nicolas et al. / MetaCitta : Deep Meta-Learning for Spatio-Temporal Prediction Across Cities and Tasks. Advances in Knowledge Discovery and Data Mining: 27th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2023, Osaka, Japan, May 25–28, 2023, Proceedings, Part IV. editor / Hisashi Kashima ; Tsuyoshi Ide ; Wen-Chih Peng. Springer Science and Business Media Deutschland GmbH, 2023. pp. 70-82 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
Download
@inproceedings{cf01519d28d6440091d1a5184cd81421,
title = "MetaCitta: Deep Meta-Learning for Spatio-Temporal Prediction Across Cities and Tasks",
abstract = "Accurate spatio-temporal prediction is essential for capturing city dynamics and planning mobility services. State-of-the-art deep spatio-temporal predictive models depend on rich and representative training data for target regions and tasks. However, the availability of such data is typically limited. Furthermore, existing predictive models fail to utilize cross-correlations across tasks and cities. In this paper, we propose MetaCitta, a novel deep meta-learning approach that addresses the critical challenges of data scarcity and model generalization. MetaCitta adopts the data from different cities and tasks in a generalizable spatio-temporal deep neural network. We propose a novel meta-learning algorithm that minimizes the discrepancy between spatio-temporal representations across tasks and cities. Our experiments with real-world data demonstrate that the proposed MetaCitta approach outperforms state-of-the-art prediction methods for zero-shot learning and pre-training plus fine-tuning. Furthermore, MetaCitta is computationally more efficient than the existing meta-learning approaches.",
keywords = "Meta-Learning, Pre-training, Spatio-Temporal Prediction",
author = "Ashutosh Sao and Simon Gottschalk and Nicolas Tempelmeier and Elena Demidova",
note = "Funding Information: This work was partially funded by the DFG, German Research Foundation (“WorldKG”, 424985896), the Federal Ministry for Economic Affairs and Climate Action (BMWK), Germany (“d-E-mand”, 01ME19009B), and DAAD, Germany (“KOALA”, 57600865). ; 27th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2023 ; Conference date: 25-05-2023 Through 28-05-2023",
year = "2023",
month = may,
day = "26",
doi = "10.1007/978-3-031-33383-5_6",
language = "English",
isbn = "9783031333828",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Science and Business Media Deutschland GmbH",
pages = "70--82",
editor = "Hisashi Kashima and Tsuyoshi Ide and Wen-Chih Peng",
booktitle = "Advances in Knowledge Discovery and Data Mining",
address = "Germany",

}

Download

TY - GEN

T1 - MetaCitta

T2 - 27th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2023

AU - Sao, Ashutosh

AU - Gottschalk, Simon

AU - Tempelmeier, Nicolas

AU - Demidova, Elena

N1 - Funding Information: This work was partially funded by the DFG, German Research Foundation (“WorldKG”, 424985896), the Federal Ministry for Economic Affairs and Climate Action (BMWK), Germany (“d-E-mand”, 01ME19009B), and DAAD, Germany (“KOALA”, 57600865).

PY - 2023/5/26

Y1 - 2023/5/26

N2 - Accurate spatio-temporal prediction is essential for capturing city dynamics and planning mobility services. State-of-the-art deep spatio-temporal predictive models depend on rich and representative training data for target regions and tasks. However, the availability of such data is typically limited. Furthermore, existing predictive models fail to utilize cross-correlations across tasks and cities. In this paper, we propose MetaCitta, a novel deep meta-learning approach that addresses the critical challenges of data scarcity and model generalization. MetaCitta adopts the data from different cities and tasks in a generalizable spatio-temporal deep neural network. We propose a novel meta-learning algorithm that minimizes the discrepancy between spatio-temporal representations across tasks and cities. Our experiments with real-world data demonstrate that the proposed MetaCitta approach outperforms state-of-the-art prediction methods for zero-shot learning and pre-training plus fine-tuning. Furthermore, MetaCitta is computationally more efficient than the existing meta-learning approaches.

AB - Accurate spatio-temporal prediction is essential for capturing city dynamics and planning mobility services. State-of-the-art deep spatio-temporal predictive models depend on rich and representative training data for target regions and tasks. However, the availability of such data is typically limited. Furthermore, existing predictive models fail to utilize cross-correlations across tasks and cities. In this paper, we propose MetaCitta, a novel deep meta-learning approach that addresses the critical challenges of data scarcity and model generalization. MetaCitta adopts the data from different cities and tasks in a generalizable spatio-temporal deep neural network. We propose a novel meta-learning algorithm that minimizes the discrepancy between spatio-temporal representations across tasks and cities. Our experiments with real-world data demonstrate that the proposed MetaCitta approach outperforms state-of-the-art prediction methods for zero-shot learning and pre-training plus fine-tuning. Furthermore, MetaCitta is computationally more efficient than the existing meta-learning approaches.

KW - Meta-Learning

KW - Pre-training

KW - Spatio-Temporal Prediction

UR - http://www.scopus.com/inward/record.url?scp=85173559157&partnerID=8YFLogxK

U2 - 10.1007/978-3-031-33383-5_6

DO - 10.1007/978-3-031-33383-5_6

M3 - Conference contribution

AN - SCOPUS:85173559157

SN - 9783031333828

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 70

EP - 82

BT - Advances in Knowledge Discovery and Data Mining

A2 - Kashima, Hisashi

A2 - Ide, Tsuyoshi

A2 - Peng, Wen-Chih

PB - Springer Science and Business Media Deutschland GmbH

Y2 - 25 May 2023 through 28 May 2023

ER -

By the same author(s)