The Fundamental Limits of Least-Privilege Learning

Theresa Stadler, Bogdan Kulynych, Michael Gastpar, Nicolas Papernot, Carmela Troncoso
Proceedings of the 41st International Conference on Machine Learning, PMLR 235:46393-46411, 2024.

Abstract

The promise of least-privilege learning – to find feature representations that are useful for a learning task but prevent inference of any sensitive information unrelated to this task – is highly appealing. However, so far this concept has only been stated informally. It thus remains an open question whether and how we can achieve this goal. In this work, we provide the first formalisation of the least-privilege principle for machine learning and characterise its feasibility. We prove that there is a fundamental trade-off between a representation’s utility for a given task and its leakage beyond the intended task: it is not possible to learn representations that have high utility for the intended task but, at the same time, prevent inference of any attribute other than the task label itself. This trade-off holds regardless of the technique used to learn the feature mappings that produce these representations. We empirically validate this result for a wide range of learning techniques, model architectures, and datasets.

Cite this Paper

BibTeX
@InProceedings{pmlr-v235-stadler24a, title = {The Fundamental Limits of Least-Privilege Learning}, author = {Stadler, Theresa and Kulynych, Bogdan and Gastpar, Michael and Papernot, Nicolas and Troncoso, Carmela}, booktitle = {Proceedings of the 41st International Conference on Machine Learning}, pages = {46393--46411}, 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/stadler24a/stadler24a.pdf}, url = {https://proceedings.mlr.press/v235/stadler24a.html}, abstract = {The promise of least-privilege learning – to find feature representations that are useful for a learning task but prevent inference of any sensitive information unrelated to this task – is highly appealing. However, so far this concept has only been stated informally. It thus remains an open question whether and how we can achieve this goal. In this work, we provide the first formalisation of the least-privilege principle for machine learning and characterise its feasibility. We prove that there is a fundamental trade-off between a representation’s utility for a given task and its leakage beyond the intended task: it is not possible to learn representations that have high utility for the intended task but, at the same time, prevent inference of any attribute other than the task label itself. This trade-off holds regardless of the technique used to learn the feature mappings that produce these representations. We empirically validate this result for a wide range of learning techniques, model architectures, and datasets.} }
Endnote
%0 Conference Paper %T The Fundamental Limits of Least-Privilege Learning %A Theresa Stadler %A Bogdan Kulynych %A Michael Gastpar %A Nicolas Papernot %A Carmela Troncoso %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-stadler24a %I PMLR %P 46393--46411 %U https://proceedings.mlr.press/v235/stadler24a.html %V 235 %X The promise of least-privilege learning – to find feature representations that are useful for a learning task but prevent inference of any sensitive information unrelated to this task – is highly appealing. However, so far this concept has only been stated informally. It thus remains an open question whether and how we can achieve this goal. In this work, we provide the first formalisation of the least-privilege principle for machine learning and characterise its feasibility. We prove that there is a fundamental trade-off between a representation’s utility for a given task and its leakage beyond the intended task: it is not possible to learn representations that have high utility for the intended task but, at the same time, prevent inference of any attribute other than the task label itself. This trade-off holds regardless of the technique used to learn the feature mappings that produce these representations. We empirically validate this result for a wide range of learning techniques, model architectures, and datasets.
APA
Stadler, T., Kulynych, B., Gastpar, M., Papernot, N. & Troncoso, C.. (2024). The Fundamental Limits of Least-Privilege Learning. Proceedings of the 41st International Conference on Machine Learning, in Proceedings of Machine Learning Research 235:46393-46411 Available from https://proceedings.mlr.press/v235/stadler24a.html.

Related Material