no code implementations • 7 Feb 2024 • Marlon Tobaben, Gauri Pradhan, Yuan He, Joonas Jälkö, Antti Honkela
We apply a state-of-the-art membership inference attack (MIA) to systematically test the practical privacy vulnerability of fine-tuning large image classification models. We focus on understanding the properties of data sets and samples that make them vulnerable to membership inference.
no code implementations • 6 Feb 2024 • Ossi Räisä, Antti Honkela
We investigate how our theory works in practice by evaluating the performance of an ensemble over many synthetic datasets for several real datasets and downstream predictors.
no code implementations • 6 Feb 2024 • Ossi Räisä, Joonas Jälkö, Antti Honkela
The remaining subsampling-induced variance decreases with larger batch sizes, so large batches reduce the effective total gradient variance.
no code implementations • 15 Dec 2023 • Rubèn Tito, Khanh Nguyen, Marlon Tobaben, Raouf Kerkouche, Mohamed Ali Souibgui, Kangsoo Jung, Lei Kang, Ernest Valveny, Antti Honkela, Mario Fritz, Dimosthenis Karatzas
We employ a federated learning scheme, that reflects the real-life distribution of documents in different businesses, and we explore the use case where the ID of the invoice issuer is the sensitive information to be protected.
1 code implementation • 9 Aug 2023 • Lukas Prediger, Joonas Jälkö, Antti Honkela, Samuel Kaski
Consider a setting where multiple parties holding sensitive data aim to collaboratively learn population level statistics, but pooling the sensitive data sets is not possible.
1 code implementation • 2 Feb 2023 • Marlon Tobaben, Aliaksandra Shysheya, John Bronskill, Andrew Paverd, Shruti Tople, Santiago Zanella-Beguelin, Richard E Turner, Antti Honkela
There has been significant recent progress in training differentially private (DP) models which achieve accuracy that approaches the best non-private models.
no code implementations • 28 Oct 2022 • Joonas Jälkö, Lukas Prediger, Antti Honkela, Samuel Kaski
Using this as prior knowledge we establish a link between the gradients of the variational parameters, and propose an efficient while simple fix for the problem to obtain a less noisy gradient estimator, which we call $\textit{aligned}$ gradients.
1 code implementation • 30 Sep 2022 • Antti Koskela, Marlon Tobaben, Antti Honkela
In order to account for the individual privacy losses in a principled manner, we need a privacy accountant for adaptive compositions of randomised mechanisms, where the loss incurred at a given data access is allowed to be smaller than the worst-case loss.
1 code implementation • 23 Sep 2022 • Mikko A. Heikkilä, Matthew Ashman, Siddharth Swaroop, Richard E. Turner, Antti Honkela
In this paper, we present differentially private partitioned variational inference, the first general framework for learning a variational approximation to a Bayesian posterior distribution in the federated learning setting while minimising the number of communication rounds and providing differential privacy guarantees for data subjects.
2 code implementations • 28 May 2022 • Ossi Räisä, Joonas Jälkö, Samuel Kaski, Antti Honkela
For example, confidence intervals become too narrow, which we demonstrate with a simple experiment.
no code implementations • 27 Oct 2021 • tejas kulkarni, Joonas Jälkö, Samuel Kaski, Antti Honkela
In recent years, local differential privacy (LDP) has emerged as a technique of choice for privacy-preserving data collection in several scenarios when the aggregator is not trustworthy.
no code implementations • 1 Jun 2021 • Antti Honkela, Laila Melkas
We achieve this by using sparse GP methodology and publishing a private variational approximation on known inducing points.
no code implementations • 1 Jun 2021 • Antti Koskela, Mikko A. Heikkilä, Antti Honkela
Shuffle model of differential privacy is a novel distributed privacy model based on a combination of local privacy mechanisms and a secure shuffler.
1 code implementation • 22 Mar 2021 • Lukas Prediger, Niki Loppi, Samuel Kaski, Antti Honkela
We present d3p, a software package designed to help fielding runtime efficient widely-applicable Bayesian inference under differential privacy guarantees.
no code implementations • 24 Feb 2021 • Antti Koskela, Antti Honkela
The recently proposed Fast Fourier Transform (FFT)-based accountant for evaluating $(\varepsilon,\delta)$-differential privacy guarantees using the privacy loss distribution formalism has been shown to give tighter bounds than commonly used methods such as R\'enyi accountants when applied to homogeneous compositions, i. e., to compositions of identical mechanisms.
no code implementations • 1 Nov 2020 • tejas kulkarni, Joonas Jälkö, Antti Koskela, Samuel Kaski, Antti Honkela
Generalized linear models (GLMs) such as logistic regression are among the most widely used arms in data analyst's repertoire and often used on sensitive datasets.
no code implementations • 19 Oct 2020 • Razane Tajeddine, Joonas Jälkö, Samuel Kaski, Antti Honkela
We modify a secure multiparty computation (MPC) framework to combine MPC with differential privacy (DP), in order to use differentially private MPC effectively to learn a probabilistic generative model under DP on such vertically partitioned data.
1 code implementation • 10 Jul 2020 • Mikko A. Heikkilä, Antti Koskela, Kana Shimizu, Samuel Kaski, Antti Honkela
In this paper we combine additively homomorphic secure summation protocols with differential privacy in the so-called cross-silo federated learning setting.
1 code implementation • 12 Jun 2020 • Antti Koskela, Joonas Jälkö, Lukas Prediger, Antti Honkela
We carry out an error analysis of the method in terms of moment bounds of the privacy loss distribution which leads to rigorous lower and upper bounds for the true $(\varepsilon,\delta)$-values.
2 code implementations • 10 Dec 2019 • Joonas Jälkö, Eemil Lagerspetz, Jari Haukka, Sasu Tarkoma, Antti Honkela, Samuel Kaski
Differential privacy allows quantifying privacy loss resulting from accessing sensitive personal data.
1 code implementation • 24 Nov 2019 • Mrinank Sharma, Michael Hutchinson, Siddharth Swaroop, Antti Honkela, Richard E. Turner
This setting is known as federated learning, in which privacy is a key concern.
1 code implementation • 7 Jun 2019 • Antti Koskela, Joonas Jälkö, Antti Honkela
The privacy loss of DP algorithms is commonly reported using $(\varepsilon,\delta)$-DP.
no code implementations • 29 Jan 2019 • Teppo Niinimäki, Mikko Heikkilä, Antti Honkela, Samuel Kaski
Differentially private learning with genomic data is challenging because it is more difficult to guarantee the privacy in high dimensions.
1 code implementation • NeurIPS 2019 • Mikko A. Heikkilä, Joonas Jälkö, Onur Dikmen, Antti Honkela
Recent developments in differentially private (DP) machine learning and DP Bayesian learning have enabled learning under strong privacy guarantees for the training data subjects.
1 code implementation • 11 Sep 2018 • Antti Koskela, Antti Honkela
We also show that it works robustly in the case of federated learning unlike commonly used optimisation methods.
1 code implementation • NeurIPS 2017 • Mikko Heikkilä, Eemil Lagerspetz, Samuel Kaski, Kana Shimizu, Sasu Tarkoma, Antti Honkela
Many applications of machine learning, for example in health care, would benefit from methods that can guarantee privacy of data subjects.
2 code implementations • 27 Oct 2016 • Joonas Jälkö, Onur Dikmen, Antti Honkela
It is built on top of doubly stochastic variational inference, a recent advance which provides a variational solution to a large class of models.
no code implementations • 7 Jun 2016 • Antti Honkela, Mrinal Das, Arttu Nieminen, Onur Dikmen, Samuel Kaski
Good personalised predictions are vitally important in precision medicine, but genomic information on which the predictions are based is also particularly sensitive, as it directly identifies the patients and hence cannot easily be anonymised.
no code implementations • 8 Mar 2016 • Otte Heinävaara, Janne Leppä-aho, Jukka Corander, Antti Honkela
Various $\ell_1$-penalised estimation methods such as graphical lasso and CLIME are widely used for sparse precision matrix estimation.