Practical Secure Aggregation for Privacy-Preserving Machine Learning

By Keith Bonawitz, Vladimir Ivanov, Ben Kreuter, Antonio Marcedone, H. Brendan McMahan, Sarvar Patel, Daniel Ramage, Aaron Segal, Karn Seth. 2017

4,229 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Federated & Distributed Learning, Privacy & Security

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 Federated & Distributed Learning

In Privacy & Security

Citations in the corpus

Advances and Open Problems in Federated Learning (2019, background): “[73], effectively handling dropping out participants at the cost of complexity of the protocol.”
Deep Models Under the GAN: Information Leakage from Collaborative Deep Learning (2017, methodology): “directly without transferring sensitive data to the company’s data centers [8, 49].”
The Distributed Discrete Gaussian Mechanism for Federated Learning with Secure Aggregation (2021, background): “…(McMahan et al., 2017; Kairouz et al., 2019), cryptographic secure aggregation (SecAgg) to prevent a server from inspecting individual user updates (Bonawitz et al., 2017; Bell et al., 2020), and differentially private stochastic gradient descent (DP-SGD) to train models with provably limited…”
Privacy Amplification by Decentralization (2020, background): “A popular approach is to resort to cryptographic primitives to securely aggregate user contributions [14, 31, 7, 10, 21].”
Differentially Private Learning Needs Better Features (or Much More Data) (2020, methodology): “Cryptography can split the trust in a central party across users (Jayaraman et al., 2018; Bonawitz et al., 2017).”
Secure Single-Server Aggregation with (Poly)Logarithmic Overhead (2020, background): “[8] uses the server as a relay that forwards encrypted and authenticated messages between clients.”
Flower: A Friendly Federated Learning Research Framework (2020, methodology): “Flower’s implementation of Secure Aggregation, named Salvia, is based on the SecAgg (Bonawitz et al., 2017) and SecAgg + (Bell et al., 2020) protocols as described in Section 3.6.”
FedML: A Research Library and Benchmark for Federated Machine Learning (2020, methodology): “Finally, we plan to provide a sample implementation of the secure aggregation algorithm [113], using the above APIs.”
Attack of the Tails: Yes, You Really Can Backdoor Federated Learning (2020, background): “, a secure version of model averaging [26].”
Backdoor Attacks on Federated Meta-Learning (2020, background): “Additional privacy threats have also been addressed in federated learning: for example, secure aggregation techniques have been proposed where users send encrypted updates that the server applies to an encrypted model [2, 3].”
Privacy in Deep Learning: A Survey (2020, methodology): “-Differential Privacy ( -DP).”
Salvaging Federated Learning by Local Adaptation (2020, background): “Secure aggregation (Bonawitz et al., 2017) prevents the global server from observing individual updates, but it also makes attacks on integrity impossible to detect and the ﬁnal model may still leak training data.”
Turbo-Aggregate: Breaking the Quadratic Aggregation Barrier in Secure Federated Learning (2020, methodology): “This approach, also known as secure aggregation [3], [4], does not sacrifice the convergence performance.”
Robust Aggregation for Federated Learning (2019, methodology): “In practice, a secure average oracle is implemented using cryptographic protocols based on secure multi-party computation [13, 17].”
Can You Really Backdoor Federated Learning? (2019, background): “Model update poisoning attacks are even harder to counter when secure aggregation (SecAgg) (7), which ensures that the server cannot inspect each user’s update, is deployed in the aggregation of local updates.”
Secure Federated Submodel Learning (2019, methodology): “only suffices for the recommendation models used in Alibaba but also can easily degenerate to a widely used vector-based representation [18], [30], by setting the number of columns d to 1.”