Upgrade to Pro — share decks privately, control downloads, hide ads and more …

JSTさきがけ「リライアブルな意思決定のための時空間因果推論モデルの研究」、「信頼されるAI」領域

koh_t
November 25, 2023
540

 JSTさきがけ「リライアブルな意思決定のための時空間因果推論モデルの研究」、「信頼されるAI」領域

2023年11月25日(土)
JSTさきがけ「信頼されるAIの基盤技術」領域 2023年度 第1回研究成果報告会
リライアブルな意思決定のための時空間因果推論モデルの研究
京都大学 竹内 孝
https://www.ml.ist.i.kyoto-u.ac.jp/koh.takeuchi/
https://www.jst.go.jp/kisoken/aip/result/event/trustworthy-ai2023/index.html

koh_t

November 25, 2023
Tweet

Transcript

  1. 4

  2. • • • 10 Koh Takeuchi, Ryo Nishida, Hisashi Kashima,

    Masaki Onishi. ” Causal Effect Estimation on Hierarchical Spatial Graph Data. ” In Proceedings of the 29th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), 2023.
  3. 11 , , , , , , , 2016, 31

    , 6 , p. AG-I_1-10, 2017
  4. 13

  5. • • • • • 14 M. Barth and K.

    Boriboonsomsin, Real-World Carbon Dioxide Impacts of Traffic Congestion. Transportation Research Record 2058, 1 (2008).
  6. • • 15 Ryu Shirakami, Toshiya Kitahara, Koh Takeuchi, Hisashi

    Kashima. “QTNet: Theory-based Queue Length Prediction for Urban Traffic. “ In Proceedings of the 29th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), 2023. この車群が到着するころに 青になるよう制御 待ち行列台数 流入交通量
  7. 16

  8. • • 18 Ryu Shirakami, Toshiya Kitahara, Koh Takeuchi, Hisashi

    Kashima. “QTNet: Theory-based Queue Length Prediction for Urban Traffic. “ In Proceedings of the 29th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), 2023. • • • • • •
  9. 19 Ryu Shirakami, Toshiya Kitahara, Koh Takeuchi, Hisashi Kashima. “QTNet:

    Theory-based Queue Length Prediction for Urban Traffic. “ In Proceedings of the 29th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), 2023. 𝑃𝑇𝐹, 𝑆𝑇𝐹 STGNN QT-layer Encoders Decoders 渋滞長 平均速度 流入交通量 補正項 平均速度 流入交通量 Inputs: 直近T’タイムスタンプの交通データ Outputs: 将来Tタイムスタンプ先までの交通データ 渋滞長
  10. • • • • 20 Weiwei Jiang, Jiayun Luo, “Graph

    neural network for traffic forecasting: A survey”, Expert Systems with Applications, Vol. 207, 2022, 1 2 3 4 1 2 3 4 1 2 3 4
  11. • • 22 Ryu Shirakami, Toshiya Kitahara, Koh Takeuchi, Hisashi

    Kashima. “QTNet: Theory-based Queue Length Prediction for Urban Traffic. “ In Proceedings of the 29th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), 2023.
  12. • ➔ 23 Ryu Shirakami, Toshiya Kitahara, Koh Takeuchi, Hisashi

    Kashima. “QTNet: Theory-based Queue Length Prediction for Urban Traffic. “ In Proceedings of the 29th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), 2023. 交通量 渋滞長 QT-layer STGNN QTNN 速度 流入交通量の増加によって 渋滞長が伸びそうだ
  13. • 24 渋滞長を 予測 信号 制 御 交通管制セン タ ー

    カ ーナビ ・ 地図アプリ 経 路 誘導
  14. 25

  15. • 27 渡り鳥※1,2の飛行軌跡データ Koh Takeuchi, Masaaki Imaizumi, Shunsuke Kanda, Yasuo

    Tabei, Keisuke Fujii, Ken Yoda, Masakazu Ishihata, Takuya Maekawa. Fréchet Kernel for Trajectory Data Analysis, Proceedings of the 29th International Conference on Advances in Geographic Information Systems (SIGSPATIAL), pp. 221-22, 2021
  16. • • 28 Keisuke Fujii, Koh Takeuchi, Atsushi Kuribayashi, Naoya

    Takeishi, Yoshinobu Kawahara, Kazuya Takeda. “Estimating counterfactual treatment outcomes over time in multi-vehicle simulation,” In Proceedings of the 30th International Conference on Advances in Geographic Information Systems (SIGSPATIAL), 2022.
  17. • • • • • • • • • •

    29 渋滞長を 予測 信号 制 御 交通管制セン タ ー カ ーナビ ・ 地図アプリ 経 路 誘導
  18. 30

  19. • • • 31 Koh Takeuchi, Hisashi Kashima, and Naonori

    Ueda. “Autoregressive Tensor Factorization for Spatio-temporal Predictions,” Proceedings of IEEE International Conference on Data Mining (ICDM), pp. 1105-1110, 2017.
  20. • 33 A crash course on graph theory: https://speakerdeck.com/tasusu/a-crash-course-on-graph-theory :

    https://www.jstage.jst.go.jp/browse/jjsai/36/4/_contents/-char/ja : https://www.jstage.jst.go.jp/article/jjsai/38/2/38_158/_article/-char/ja/
  21. • • • • 37 Weiwei Jiang, Jiayun Luo, “Graph

    neural network for traffic forecasting: A survey”, Expert Systems with Applications, Vol. 207, 2022, 1 2 3 4 1 2 3 4 1 2 3 4
  22. • • 38 Zonghan Wu et al. "Graph WaveNet for

    Deep Spatial-Temporal Graph Modeling,” In Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence (IJCAI), 2019. Lei Bai et al. “Adaptive Graph Convolutional Recurrent Network for Traffic Forecasting,” In Proceediongs of NeurIPS, 2020. 1 2 3 4 1 2 3 4
  23. • X Z Y • • 41 Koh Takeuchi, Ryo

    Nishida, Hisashi Kashima, Masaki Onishi. ” Grab the reins of crowds: Estimating the effects of crowd movement guidance using causal inference. ” Proceedings of the 20th International Conference on Autonomous Agents and MultiAgent Systems (AAMAS), 2021.
  24. 42 Xn ∈ 0,1 𝑀 𝑛 Ck ∈ 0,1 𝐾

    An G Yt (t=1, …, T ) Zn An Xn N G Ck K Yt T
  25. • • 46 Koh Takeuchi, Ryo Nishida, Hisashi Kashima, Masaki

    Onishi. ” Causal Effect Estimation on Hierarchical Spatial Graph Data. ” In Proceedings of the 29th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), 2023.
  26. 階層的な空間グラフデータからの介入効果推定 • 48 Koh Takeuchi, Ryo Nishida, Hisashi Kashima, Masaki

    Onishi. ” Causal Effect Estimation on Hierarchical Spatial Graph Data. ” In Proceedings of the 29th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), 2023.
  27. 49

  28. • • • 50 F. D. Johansson et al. “Learning

    Representations for Counterfactual Inference,” ICML2016. U. Shalit et al. "Estimating individual treatment effect: generalization bounds and algorithms." ICML, 2017. X C f Φ g Y
  29. • • 51 F. D. Johansson et al. “Learning Representations

    for Counterfactual Inference,” ICML2016. U. Shalit et al. "Estimating individual treatment effect: generalization bounds and algorithms." ICML, 2017. X C f Φ g Y IPM 正則化モデルの損失関数は、 の期待値の上限と証明
  30. • 反事実回帰モデル • 共変量 から抽出した特徴量 の分布が介入と • 正則化モデルの損失関数が、 の期待値の上限となると証明 52

    F. D. Johansson et al. “Learning Representations for Counterfactual Inference,” ICML2016. U. Shalit et al. "Estimating individual treatment effect: generalization bounds and algorithms." ICML, 2017. X C f Φ g Y IPM
  31. • Queueing-Theory layer (QT-Layer) • QT-Layer 3 55 Ryu Shirakami,

    Toshiya Kitahara, Koh Takeuchi, Hisashi Kashima. “QTNet: Theory-based Queue Length Prediction for Urban Traffic. “ In Proceedings of the 29th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), 2023.