Causal Action Influence Aware Counterfactual Data Augmentation

Núria Armengol Urpı́, Marco Bagatella, Marin Vlastelica, Georg Martius
Proceedings of the 41st International Conference on Machine Learning, PMLR 235:1709-1729, 2024.

Abstract

Offline data are both valuable and practical resources for teaching robots complex behaviors. Ideally, learning agents should not be constrained by the scarcity of available demonstrations, but rather generalize beyond the training distribution. However, the complexity of real-world scenarios typically requires huge amounts of data to prevent neural network policies from picking up on spurious correlations and learning non-causal relationships. We propose CAIAC, a data augmentation method that can create feasible synthetic transitions from a fixed dataset without having access to online environment interactions. By utilizing principled methods for quantifying causal influence, we are able to perform counterfactual reasoning by swapping $\textit{action}$-unaffected parts of the state-space between independent trajectories in the dataset. We empirically show that this leads to a substantial increase in robustness of offline learning algorithms against distributional shift.

Cite this Paper

BibTeX
@InProceedings{pmlr-v235-armengol-urpi-24a, title = {Causal Action Influence Aware Counterfactual Data Augmentation}, author = {Armengol Urp\'{\i}, N\'{u}ria and Bagatella, Marco and Vlastelica, Marin and Martius, Georg}, booktitle = {Proceedings of the 41st International Conference on Machine Learning}, pages = {1709--1729}, year = {2024}, editor = {Salakhutdinov, Ruslan and Kolter, Zico and Heller, Katherine and Weller, Adrian and Oliver, Nuria and Scarlett, Jonathan and Berkenkamp, Felix}, volume = {235}, series = {Proceedings of Machine Learning Research}, month = {21--27 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v235/main/assets/armengol-urpi-24a/armengol-urpi-24a.pdf}, url = {https://proceedings.mlr.press/v235/armengol-urpi-24a.html}, abstract = {Offline data are both valuable and practical resources for teaching robots complex behaviors. Ideally, learning agents should not be constrained by the scarcity of available demonstrations, but rather generalize beyond the training distribution. However, the complexity of real-world scenarios typically requires huge amounts of data to prevent neural network policies from picking up on spurious correlations and learning non-causal relationships. We propose CAIAC, a data augmentation method that can create feasible synthetic transitions from a fixed dataset without having access to online environment interactions. By utilizing principled methods for quantifying causal influence, we are able to perform counterfactual reasoning by swapping $\textit{action}$-unaffected parts of the state-space between independent trajectories in the dataset. We empirically show that this leads to a substantial increase in robustness of offline learning algorithms against distributional shift.} }
Endnote
%0 Conference Paper %T Causal Action Influence Aware Counterfactual Data Augmentation %A Núria Armengol Urpı́ %A Marco Bagatella %A Marin Vlastelica %A Georg Martius %B Proceedings of the 41st International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2024 %E Ruslan Salakhutdinov %E Zico Kolter %E Katherine Heller %E Adrian Weller %E Nuria Oliver %E Jonathan Scarlett %E Felix Berkenkamp %F pmlr-v235-armengol-urpi-24a %I PMLR %P 1709--1729 %U https://proceedings.mlr.press/v235/armengol-urpi-24a.html %V 235 %X Offline data are both valuable and practical resources for teaching robots complex behaviors. Ideally, learning agents should not be constrained by the scarcity of available demonstrations, but rather generalize beyond the training distribution. However, the complexity of real-world scenarios typically requires huge amounts of data to prevent neural network policies from picking up on spurious correlations and learning non-causal relationships. We propose CAIAC, a data augmentation method that can create feasible synthetic transitions from a fixed dataset without having access to online environment interactions. By utilizing principled methods for quantifying causal influence, we are able to perform counterfactual reasoning by swapping $\textit{action}$-unaffected parts of the state-space between independent trajectories in the dataset. We empirically show that this leads to a substantial increase in robustness of offline learning algorithms against distributional shift.
APA
Armengol Urpı́, N., Bagatella, M., Vlastelica, M. & Martius, G.. (2024). Causal Action Influence Aware Counterfactual Data Augmentation. Proceedings of the 41st International Conference on Machine Learning, in Proceedings of Machine Learning Research 235:1709-1729 Available from https://proceedings.mlr.press/v235/armengol-urpi-24a.html.

Related Material