c ○ 2011 Society for Industrial and Applied Mathematics WHAT CAN WE LEARN PRIVATELY? ∗

Subfields: Privacy & Security

Citations in the corpus

• Advances and Open Problems in Federated Learning (2019, methodology): “2, achieving LDP while maintaining utility can be difficult [229, 388].”
• Deep Models Under the GAN: Information Leakage from Collaborative Deep Learning (2017, background): “Differential privacy plays an important role in deep learning as shown in [1, 38, 75, 77].”
• Practical and Private (Deep) Learning without Sampling or Shuffling (2021, methodology): “This improvement is significant because it distinguishes centrally private OCO from locally private (Warner, 1965; Evfimievski et al., 2003; Kasiviswanathan et al., 2008) OCO4.”
• The Distributed Discrete Gaussian Mechanism for Federated Learning with Secure Aggregation (2021, methodology): “The privacy guarantees ε we report rely on privacy amplification via sampling (Kasiviswanathan et al., 2011; Bassily et al., 2014; Abadi et al., 2016), which is necessary to obtain reasonable privacy-accuracy tradeoffs”
• DISCO: Dynamic and Invariant Sensitive Channel Obfuscation for deep neural networks (2020, background): “These mechanisms are sometimes built under the framework of differential privacy [16] or its derivatives [64, 65].”
• Privacy Amplification by Decentralization (2020, background): “The strongest model is local differential privacy (LDP) (Kasiviswanathan et al., 2008; Duchi et al., 2013), where each participant (user) does not trust anyone and aims to protect against an adversary that can observe ev-a r X i v : 2012 .”
• Learning discrete distributions: user vs item-level privacy (2020, background): “Differential privacy is studied in two scenarios, local differential privacy [Kasiviswanathan et al., 2011, Duchi et al., 2013], and global differential privacy [Dwork et al., 2006].”
• Breaking the Communication-Privacy-Accuracy Trilemma (2020, background): “In particular, local differential privacy (LDP) [3], [4], [5], [6] is a widely adopted privacy paradigm, which guarantees that the outcome from a privatization mechanism will not release too much individual information statistically.”
• Privacy Amplification via Random Check-Ins (2020, background): “Source of Privacy Amplification ε for Central DP None [14, 32] ε0 Uniform subsampling [25, 9, 1] O (e0/ √ n) Shuffling [19] O ( e0/ √ n )”
• Private Counting from Anonymous Messages: Near-Optimal Accuracy with Vanishing Communication Overhead (2020, background): “To circumvent this limitation, several works have studied the local model of DP (Kasiviswanathan et al., 2008) (also (Warner, 1965)), which enforces the more stringent constraint that each message sent from a user device to the server is private.”
• Private Summation in the Multi-Message Shuffle Model (2020, neutral): “The local model of differential privacy [19] considers a subclass of all DP mechanisms which operate under the assumption that each user applies randomness to privatize their own data before sending it to (a potentially untrusted) aggregator for analysis.”
• Pure Differentially Private Summation from Anonymous Messages (2020, methodology): “To remedy this, the more appealing local model of DP (denoted DP local ) [KLN + 08] (also [War65]) requires the messages sent by each user to the analyst to be private.”
• The power of synergy in differential privacy: Combining a small curator with local randomizers (2019, methodology): “The local model of di ff erential privacy [66, 43] has recently emerged as one of the promising approaches for achieving the goals of enabling data-driven innovation while preserving a rigorous notion of privacy for individuals that also addresses the above challenges.”
• Secure Federated Submodel Learning (2019, background): “Due to its intriguing security properties, local differential privacy for various population statistics has recently received significant industrial deployments (e.g., in Google [47], [48], Apple [49], and Microsoft [50]), as well as lasting academic attention [51]–[61].”
• Differentially Private Learning with Adaptive Clipping (2019, neutral): “Privacy amplification via subsampling was introduced in (Kasiviswanathan et al., 2011).”
• The Privacy Blanket of the Shuffle Model (2019, methodology): “These include amplification by subsampling [22] and by iteration [16], and the recent seminal work on amplification via shuffling by Erlingsson et al.”
• Federated Heavy Hitters Discovery with Differential Privacy (2019, methodology): “, assumes that the users do not have any trust in the service provider (Kasiviswanathan et al., 2011).”