---
OA_place: repository
OA_type: green
_id: '19509'
abstract:
- lang: eng
text: "Machine Learning and AI have the potential to transform data-driven scientific
discovery,\r\nenabling accurate predictions for several scientific phenomena.
As many scientific questions\r\nare inherently causal, this paper looks at the
causal inference task of treatment effect\r\nestimation, where the outcome of
interest is recorded in high-dimensional observations\r\nin a Randomized Controlled
Trial (RCT). Despite being the simplest possible causal setting\r\nand a perfect
fit for deep learning, we theoretically find that many common choices in the\r\nliterature
may lead to biased estimates. To test the practical impact of these considerations,\r\nwe
recorded ISTAnt, the first real-world benchmark for causal inference downstream
tasks\r\non high-dimensional observations as an RCT studying how garden ants (Lasius
neglectus)\r\nrespond to microparticles applied onto their colony members by hygienic
grooming.\r\nComparing 6 480 models fine-tuned from state-of-the-art visual backbones,
we find that\r\nthe sampling and modeling choices significantly affect the accuracy
of the causal estimate,\r\nand that classification accuracy is not a proxy thereof.
We further validated the analysis,\r\nrepeating it on a synthetically generated
visual data set controlling the causal model. Our\r\nresults suggest that future
benchmarks should carefully consider real downstream scientific\r\nquestions,
especially causal ones. Further, we highlight guidelines for representation\r\nlearning
methods to help answer causal questions in the sciences.\r\nCode: https://github.com/CausalLearningAI/ISTAnt\r\nData:
https://doi.org/10.6084/m9.figshare.26484934.v2"
acknowledgement: We thank Piersilvio De Bartolomeis, and the full Causal Learning
and Artificial Intelligence (CLAI) group at ISTA for the extremely helpful discussions.
Riccardo Cadei was supported by a Google Research Scholar Award and a Google Initiated
Gift to Francesco Locatello. We thank the Social Immunity team at ISTA, particularly
Michaela Hönigsberger and Wilfrid Jean Louis, for supporting the ecological experiment
and Farnaz Beikzadeh Abbasi, Luisa Fiebig and Martin Estermann for annotating ant
behavior in ISTAnt.
alternative_title:
- Advances in Neural Information Processing Systems
article_processing_charge: No
arxiv: 1
author:
- first_name: Riccardo
full_name: Cadei, Riccardo
id: 0fa8b76f-72f0-11ef-b75a-a5da96e5ad6b
last_name: Cadei
- first_name: Lukas
full_name: Lindorfer, Lukas
id: 85f0e6d3-06b3-11ec-8982-8c5049fa4455
last_name: Lindorfer
- first_name: Sylvia
full_name: Cremer, Sylvia
id: 2F64EC8C-F248-11E8-B48F-1D18A9856A87
last_name: Cremer
orcid: 0000-0002-2193-3868
- first_name: Cordelia
full_name: Schmid, Cordelia
last_name: Schmid
- first_name: Francesco
full_name: Locatello, Francesco
id: 26cfd52f-2483-11ee-8040-88983bcc06d4
last_name: Locatello
orcid: 0000-0002-4850-0683
citation:
ama: 'Cadei R, Lindorfer L, Cremer S, Schmid C, Locatello F. Smoke and mirrors in
causal downstream tasks. In: 38th Conference on Neural Information Processing
Systems. Vol 37. Neural Information Processing Systems Foundation; 2024. doi:10.52202/079017-0820'
apa: 'Cadei, R., Lindorfer, L., Cremer, S., Schmid, C., & Locatello, F. (2024).
Smoke and mirrors in causal downstream tasks. In 38th Conference on Neural
Information Processing Systems (Vol. 37). Vancouver, Canada: Neural Information
Processing Systems Foundation. https://doi.org/10.52202/079017-0820'
chicago: Cadei, Riccardo, Lukas Lindorfer, Sylvia Cremer, Cordelia Schmid, and Francesco
Locatello. “Smoke and Mirrors in Causal Downstream Tasks.” In 38th Conference
on Neural Information Processing Systems, Vol. 37. Neural Information Processing
Systems Foundation, 2024. https://doi.org/10.52202/079017-0820.
ieee: R. Cadei, L. Lindorfer, S. Cremer, C. Schmid, and F. Locatello, “Smoke and
mirrors in causal downstream tasks,” in 38th Conference on Neural Information
Processing Systems, Vancouver, Canada, 2024, vol. 37.
ista: 'Cadei R, Lindorfer L, Cremer S, Schmid C, Locatello F. 2024. Smoke and mirrors
in causal downstream tasks. 38th Conference on Neural Information Processing Systems.
NeurIPS: Neural Information Processing Systems, Advances in Neural Information
Processing Systems, vol. 37.'
mla: Cadei, Riccardo, et al. “Smoke and Mirrors in Causal Downstream Tasks.” 38th
Conference on Neural Information Processing Systems, vol. 37, Neural Information
Processing Systems Foundation, 2024, doi:10.52202/079017-0820.
short: R. Cadei, L. Lindorfer, S. Cremer, C. Schmid, F. Locatello, in:, 38th Conference
on Neural Information Processing Systems, Neural Information Processing Systems
Foundation, 2024.
conference:
end_date: 2024-12-15
location: Vancouver, Canada
name: 'NeurIPS: Neural Information Processing Systems'
start_date: 2024-12-09
das_tickbox: '1'
dataavailabilitystatement: "Code: https://github.com/CausalLearningAI/ISTAnt\r\nData:
https://doi.org/10.6084/m9.figshare.26484934.v2"
date_created: 2025-04-06T22:01:32Z
date_published: 2024-12-20T00:00:00Z
date_updated: 2026-10-21T06:45:33Z
day: '20'
department:
- _id: SyCr
- _id: FrLo
doi: 10.52202/079017-0820
external_id:
arxiv:
- '2405.17151'
fulldoi: https://doi.org/10.52202/079017-0820
intvolume: ' 37'
language:
- iso: eng
main_file_link:
- open_access: '1'
url: https://doi.org/10.48550/arXiv.2405.17151
month: '12'
oa: 1
oa_version: Preprint
publication: 38th Conference on Neural Information Processing Systems
publication_identifier:
eissn:
- 1049-5258
publication_status: published
publisher: Neural Information Processing Systems Foundation
quality_controlled: '1'
related_material:
link:
- relation: software
url: https://github.com/CausalLearningAI/ISTAnt
record:
- id: '18847'
relation: earlier_version
status: public
researchdata_availability: yes
scopus_import: '1'
status: public
supplementarymaterial: yes
title: Smoke and mirrors in causal downstream tasks
type: conference
user_id: 2DF688A6-F248-11E8-B48F-1D18A9856A87
volume: 37
year: '2024'
...