Standard
Learning Geometric Representations of Objects via Interaction. / Reichlin, Alfredo; Marchetti, Giovanni Luca; Yin, Hang; Varava, Anastasiia; Kragic, Danica.
Machine Learning and Knowledge Discovery in Databases: Research Track - European Conference, ECML PKDD 2023, Proceedings. ed. / Danai Koutra; Claudia Plant; Manuel Gomez Rodriguez; Elena Baralis; Francesco Bonchi. Vol. 4 Springer, 2023. p. 629-644 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Vol. 14172 LNAI).
Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review
Harvard
Reichlin, A, Marchetti, GL
, Yin, H, Varava, A & Kragic, D 2023,
Learning Geometric Representations of Objects via Interaction. in D Koutra, C Plant, M Gomez Rodriguez, E Baralis & F Bonchi (eds),
Machine Learning and Knowledge Discovery in Databases: Research Track - European Conference, ECML PKDD 2023, Proceedings. vol. 4, Springer, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 14172 LNAI, pp. 629-644, European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2023, Turin, Italy,
18/09/2023.
https://doi.org/10.1007/978-3-031-43421-1_37APA
Reichlin, A., Marchetti, G. L.
, Yin, H., Varava, A., & Kragic, D. (2023).
Learning Geometric Representations of Objects via Interaction. In D. Koutra, C. Plant, M. Gomez Rodriguez, E. Baralis, & F. Bonchi (Eds.),
Machine Learning and Knowledge Discovery in Databases: Research Track - European Conference, ECML PKDD 2023, Proceedings (Vol. 4, pp. 629-644). Springer. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) Vol. 14172 LNAI
https://doi.org/10.1007/978-3-031-43421-1_37Vancouver
Reichlin A, Marchetti GL
, Yin H, Varava A, Kragic D.
Learning Geometric Representations of Objects via Interaction. In Koutra D, Plant C, Gomez Rodriguez M, Baralis E, Bonchi F, editors, Machine Learning and Knowledge Discovery in Databases: Research Track - European Conference, ECML PKDD 2023, Proceedings. Vol. 4. Springer. 2023. p. 629-644. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Vol. 14172 LNAI).
https://doi.org/10.1007/978-3-031-43421-1_37Author
Reichlin, Alfredo ; Marchetti, Giovanni Luca ; Yin, Hang ; Varava, Anastasiia ; Kragic, Danica. / Learning Geometric Representations of Objects via Interaction. Machine Learning and Knowledge Discovery in Databases: Research Track - European Conference, ECML PKDD 2023, Proceedings. editor / Danai Koutra ; Claudia Plant ; Manuel Gomez Rodriguez ; Elena Baralis ; Francesco Bonchi. Vol. 4 Springer, 2023. pp. 629-644 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Vol. 14172 LNAI).
Bibtex
@inproceedings{57df6cbad407476c8b08c6e9c5772b4e,
title = "Learning Geometric Representations of Objects via Interaction",
abstract = "We address the problem of learning representations from observations of a scene involving an agent and an external object the agent interacts with. To this end, we propose a representation learning framework extracting the location in physical space of both the agent and the object from unstructured observations of arbitrary nature. Our framework relies on the actions performed by the agent as the only source of supervision, while assuming that the object is displaced by the agent via unknown dynamics. We provide a theoretical foundation and formally prove that an ideal learner is guaranteed to infer an isometric representation, disentangling the agent from the object and correctly extracting their locations. We evaluate empirically our framework on a variety of scenarios, showing that it outperforms vision-based approaches such as a state-of-the-art keypoint extractor. We moreover demonstrate how the extracted representations enable the agent to solve downstream tasks via reinforcement learning in an efficient manner.",
keywords = "Equivariance, Interaction, Representation Learning",
author = "Alfredo Reichlin and Marchetti, {Giovanni Luca} and Hang Yin and Anastasiia Varava and Danica Kragic",
note = "Funding Information: Acknowledgements. This work was supported by the Swedish Research Council, the Knut and Alice Wallenberg Foundation, the European Research Council (ERC-BIRD-884807) and the European Horizon 2020 CANOPIES project. Hang Yin would like to acknolwedge the support by the Pioneer Centre for AI, DNRF grant number P1. Publisher Copyright: {\textcopyright} 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.; European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2023 ; Conference date: 18-09-2023 Through 22-09-2023",
year = "2023",
doi = "10.1007/978-3-031-43421-1_37",
language = "English",
isbn = "9783031434204",
volume = "4",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer",
pages = "629--644",
editor = "Danai Koutra and Claudia Plant and {Gomez Rodriguez}, Manuel and Elena Baralis and Francesco Bonchi",
booktitle = "Machine Learning and Knowledge Discovery in Databases",
address = "Switzerland",
}
RIS
TY - GEN
T1 - Learning Geometric Representations of Objects via Interaction
AU - Reichlin, Alfredo
AU - Marchetti, Giovanni Luca
AU - Yin, Hang
AU - Varava, Anastasiia
AU - Kragic, Danica
N1 - Funding Information:
Acknowledgements. This work was supported by the Swedish Research Council, the Knut and Alice Wallenberg Foundation, the European Research Council (ERC-BIRD-884807) and the European Horizon 2020 CANOPIES project. Hang Yin would like to acknolwedge the support by the Pioneer Centre for AI, DNRF grant number P1.
Publisher Copyright:
© 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.
PY - 2023
Y1 - 2023
N2 - We address the problem of learning representations from observations of a scene involving an agent and an external object the agent interacts with. To this end, we propose a representation learning framework extracting the location in physical space of both the agent and the object from unstructured observations of arbitrary nature. Our framework relies on the actions performed by the agent as the only source of supervision, while assuming that the object is displaced by the agent via unknown dynamics. We provide a theoretical foundation and formally prove that an ideal learner is guaranteed to infer an isometric representation, disentangling the agent from the object and correctly extracting their locations. We evaluate empirically our framework on a variety of scenarios, showing that it outperforms vision-based approaches such as a state-of-the-art keypoint extractor. We moreover demonstrate how the extracted representations enable the agent to solve downstream tasks via reinforcement learning in an efficient manner.
AB - We address the problem of learning representations from observations of a scene involving an agent and an external object the agent interacts with. To this end, we propose a representation learning framework extracting the location in physical space of both the agent and the object from unstructured observations of arbitrary nature. Our framework relies on the actions performed by the agent as the only source of supervision, while assuming that the object is displaced by the agent via unknown dynamics. We provide a theoretical foundation and formally prove that an ideal learner is guaranteed to infer an isometric representation, disentangling the agent from the object and correctly extracting their locations. We evaluate empirically our framework on a variety of scenarios, showing that it outperforms vision-based approaches such as a state-of-the-art keypoint extractor. We moreover demonstrate how the extracted representations enable the agent to solve downstream tasks via reinforcement learning in an efficient manner.
KW - Equivariance
KW - Interaction
KW - Representation Learning
U2 - 10.1007/978-3-031-43421-1_37
DO - 10.1007/978-3-031-43421-1_37
M3 - Article in proceedings
AN - SCOPUS:85174436596
SN - 9783031434204
VL - 4
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 629
EP - 644
BT - Machine Learning and Knowledge Discovery in Databases
A2 - Koutra, Danai
A2 - Plant, Claudia
A2 - Gomez Rodriguez, Manuel
A2 - Baralis, Elena
A2 - Bonchi, Francesco
PB - Springer
T2 - European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2023
Y2 - 18 September 2023 through 22 September 2023
ER -
