Differentiable Multi-Target Causal Bayesian Experimental Design

Panagiotis Tigas, Yashas Annadani, Desi R. Ivanova, Andrew Jesson, Yarin Gal, Adam Foster, Stefan Bauer
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:34263-34279, 2023.

Abstract

We introduce a gradient-based approach for the problem of Bayesian optimal experimental design to learn causal models in a batch setting — a critical component for causal discovery from finite data where interventions can be costly or risky. Existing methods rely on greedy approximations to construct a batch of experiments while using black-box methods to optimize over a single target-state pair to intervene with. In this work, we completely dispose of the black-box optimization techniques and greedy heuristics and instead propose a conceptually simple end-to-end gradient-based optimization procedure to acquire a set of optimal intervention target-value pairs. Such a procedure enables parameterization of the design space to efficiently optimize over a batch of multi-target-state interventions, a setting which has hitherto not been explored due to its complexity. We demonstrate that our proposed method outperforms baselines and existing acquisition strategies in both single-target and multi-target settings across a number of synthetic datasets.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-tigas23a, title = {Differentiable Multi-Target Causal {B}ayesian Experimental Design}, author = {Tigas, Panagiotis and Annadani, Yashas and Ivanova, Desi R. and Jesson, Andrew and Gal, Yarin and Foster, Adam and Bauer, Stefan}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {34263--34279}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/tigas23a/tigas23a.pdf}, url = {https://proceedings.mlr.press/v202/tigas23a.html}, abstract = {We introduce a gradient-based approach for the problem of Bayesian optimal experimental design to learn causal models in a batch setting — a critical component for causal discovery from finite data where interventions can be costly or risky. Existing methods rely on greedy approximations to construct a batch of experiments while using black-box methods to optimize over a single target-state pair to intervene with. In this work, we completely dispose of the black-box optimization techniques and greedy heuristics and instead propose a conceptually simple end-to-end gradient-based optimization procedure to acquire a set of optimal intervention target-value pairs. Such a procedure enables parameterization of the design space to efficiently optimize over a batch of multi-target-state interventions, a setting which has hitherto not been explored due to its complexity. We demonstrate that our proposed method outperforms baselines and existing acquisition strategies in both single-target and multi-target settings across a number of synthetic datasets.} }
Endnote
%0 Conference Paper %T Differentiable Multi-Target Causal Bayesian Experimental Design %A Panagiotis Tigas %A Yashas Annadani %A Desi R. Ivanova %A Andrew Jesson %A Yarin Gal %A Adam Foster %A Stefan Bauer %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-tigas23a %I PMLR %P 34263--34279 %U https://proceedings.mlr.press/v202/tigas23a.html %V 202 %X We introduce a gradient-based approach for the problem of Bayesian optimal experimental design to learn causal models in a batch setting — a critical component for causal discovery from finite data where interventions can be costly or risky. Existing methods rely on greedy approximations to construct a batch of experiments while using black-box methods to optimize over a single target-state pair to intervene with. In this work, we completely dispose of the black-box optimization techniques and greedy heuristics and instead propose a conceptually simple end-to-end gradient-based optimization procedure to acquire a set of optimal intervention target-value pairs. Such a procedure enables parameterization of the design space to efficiently optimize over a batch of multi-target-state interventions, a setting which has hitherto not been explored due to its complexity. We demonstrate that our proposed method outperforms baselines and existing acquisition strategies in both single-target and multi-target settings across a number of synthetic datasets.
APA
Tigas, P., Annadani, Y., Ivanova, D.R., Jesson, A., Gal, Y., Foster, A. & Bauer, S.. (2023). Differentiable Multi-Target Causal Bayesian Experimental Design. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:34263-34279 Available from https://proceedings.mlr.press/v202/tigas23a.html.

Related Material