Differential Privacy and Machine Learning: a Survey and Review

By Zhanglong Ji, Zachary Chase Lipton, Charles Peter Elkan. 2014

294 citations recorded by the source corpus.

The objective of machine learning is to extract useful information from data, while privacy is preserved by concealing information. Thus it seems hard to reconcile these competing interests. However, they frequently must be balanced when mining sensitive data. For example, medical research represents an important application where it is necessary both to extract useful information and protect patient privacy. One way to resolve the conflict is to extract general characteristics of whole populations without disclosing the private information of individuals. In this paper, we consider differential privacy, one of the most popular and powerful definitions of privacy. We explore the interplay between machine learning and differential privacy, namely privacy-preserving machine learning algorithms and learning-based data release mechanisms. We also describe some theoretical results that address what can be learned differentially privately and upper bounds of loss functions for differentially private algorithms. Finally, we present some open questions, including how to incorporate public data, how to deal with missing data in private datasets, and whether, as the number of observed samples grows arbitrarily large, differentially private machine learning algorithms can be achieved at no cost to utility as compared to corresponding non-differentially private algorithms.

Subfields: Artificial Intelligence, Computer Science Applications, Sociology and Political Science

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Science Applications

In Sociology and Political Science

Citations in the corpus

Differential Privacy for Growing Databases (2018, background): “The theoretical computer science community has been proliﬁc in designing diﬀerentially private algorithms that provide accuracy guarantees for a wide variety of machine learning problems (see [JLE14] for a survey).”
Towards Practical Differentially Private Convex Optimization (2019, background): “A broad survey of works in DP machine learning has been provided in [37].”
Trustworthy AI: From Principles to Practices (2021, unknown): “Data Preprocessing Anomaly Detection [70, 81, 257, 316] Data Anonymozation [14, 37, 232, 232] Differential Privacy [107, 110, 177]”
Deep Private-Feature Extraction (2018, methodology): “Answering to complex queries by combining simple queries is the way various learning models, such as Principal Component Analysis and k-means, can be made differentially private (see the surveys by [26] and [27]).”
Property Testing For Differential Privacy (2018, background): “An introduction can be found in Dwork and Roth (2014) and more in depth surveys can be found in Vadhan (2016), Dwork (2008), Ji et al. (2014).”
Algorithms that remember: model inversion attacks and data protection law (2018, background): “’s data alone. Differentially private approaches to machine learning work in a similar way: models lacking data relating to a particular individual function extremely similarly to those containing it [45]. This provides intrinsic protection against a great deal of model inversion attacks. Theoretica”
Differential privacy for eye-tracking data (2019, methodology): “Relevant to eye tracking are the works that have applied differential privacy definitions to machine learning [2; 15] and time-series analysis [10; 29].”
When Machine Learning Meets Privacy (2020, background): “Although there are several surveys on this topic [1, 64, 85, 177], The focus has been on a certain type of ML model or specific methods.”
Differential Privacy Techniques for Cyber Physical Systems: A Survey (2018, background): “To tackle this issue, researchers have started working over merging differential privacy data perturbation technique with machine learning algorithms [35], [343], [344].”