Causal learning | Mark Edmonds

Built a virtual “escape room” to test causal generalization: surface-level features change from room to room, while every room is governed by a common abstract causal structure (a series of levers) that describes the actions required to “unlock” it.
Ran human subject experiments verifying that human learners can discover the correct abstract causal structure.
Built a hierarchical Bayesian model that performs similarly to human learners. The causal model solved the escape room while seven state-of-the-art model-free reinforcement learning algorithms failed.
Result: both structural abstraction and feature generalization are critical for transfer learning and generalization.

Collaborators: Yixin Zhu, James Kubricht, Brandon Rothrock (NASA JPL), Hongjing Lu, Song-Chun Zhu

Publications

AAAI

Theory-based Causal Transfer: Integrating Instance-level Induction and Abstract-level Structure Learning

M. Edmonds, X. Ma, S. Qi, Y. Zhu, H. Lu, and S.-C. Zhu

Proceedings of the AAAI Conference on Artificial Intelligence (AAAI), 2020Oral

PDF Supplement Project page Code

BibTeX

@inproceedings{edmonds2020theory,
  title = {Theory-based Causal Transfer: Integrating Instance-level Induction and Abstract-level Structure Learning},
  author = {Edmonds, M. and Ma, X. and Qi, S. and Zhu, Y. and Lu, H. and Zhu, S.-C.},
  booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence (AAAI)},
  year = {2020}
}

CogSci

Decomposing Human Causal Learning: Bottom-up Associative Learning and Top-down Schema Reasoning

M. Edmonds, S. Qi, Y. Zhu, J. Kubricht, S.-C. Zhu, and H. Lu

41st Annual Meeting of the Cognitive Science Society (CogSci), 2019

PDF Project page Video

BibTeX

@inproceedings{edmonds2019decomposing,
  title = {Decomposing Human Causal Learning: Bottom-up Associative Learning and Top-down Schema Reasoning},
  author = {Edmonds, M. and Qi, S. and Zhu, Y. and Kubricht, J. and Zhu, S.-C. and Lu, H.},
  booktitle = {41st Annual Meeting of the Cognitive Science Society (CogSci)},
  year = {2019}
}

CogSci

Human Causal Transfer: Challenges for Deep Reinforcement Learning

M. Edmonds*, J. Kubricht*, C. Summers, Y. Zhu, B. Rothrock, S.-C. Zhu, and H. Lu

40th Annual Meeting of the Cognitive Science Society (CogSci), 2018Oral

PDF Project page Code Video

BibTeX

@inproceedings{edmonds2018human,
  title = {Human Causal Transfer: Challenges for Deep Reinforcement Learning},
  author = {Edmonds, M. and Kubricht, J. and Summers, C. and Zhu, Y. and Rothrock, B. and Zhu, S.-C. and Lu, H.},
  booktitle = {40th Annual Meeting of the Cognitive Science Society (CogSci)},
  year = {2018}
}

CVPR

ACRE: Abstract Causal REasoning Beyond Covariation

C. Zhang, B. Jia, M. Edmonds, S.-C. Zhu, and Y. Zhu

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021

PDF Supplement Project page Code

BibTeX

@inproceedings{zhang2021acre,
  title = {ACRE: Abstract Causal REasoning Beyond Covariation},
  author = {Zhang, C. and Jia, B. and Edmonds, M. and Zhu, S.-C. and Zhu, Y.},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year = {2021}
}