• Corpus ID: 202542794

Privacy-Preserving Bandits

@article{Malekzadeh2020PrivacyPreservingB,
  title={Privacy-Preserving Bandits},
  author={M. Malekzadeh and Dimitrios Athanasakis and Hamed Haddadi and Benjamin Livshits},
  journal={ArXiv},
  year={2020},
  volume={abs/1909.04421}
}
Contextual bandit algorithms~(CBAs) often rely on personal data to provide recommendations. Centralized CBA agents utilize potentially sensitive data from recent interactions to provide personalization to end-users. Keeping the sensitive data locally, by running a local agent on the user's device, protects the user's privacy, however, the agent requires longer to produce useful recommendations, as it does not leverage feedback from other users. This paper proposes a technique we call Privacy… 
2 Citations

Figures from this paper

Differentially Private Federated Combinatorial Bandits with Constraints
TLDR
This work investigates a group of agents working concurrently to solve similar combinatorial bandit problems while maintaining quality constraints, and proposes a Privacy-preserving Federated Combinatorial Bandit algorithm, P-FCB.
Multi-Armed Bandits with Local Differential Privacy
TLDR
This paper proves a lower bound and proposes algorithms whose regret upper bounds match the lower bound up to constant factors and adopts differential privacy.
Local Differential Privacy for Regret Minimization in Reinforcement Learning
TLDR
A lower bound for regret minimization in finite-horizon MDPs with LDP guarantees is established which shows that guaranteeing privacy has a multiplicative effect on the regret.
Federated Bandit
TLDR
Federated Bandit, a decentralized Multi-Armed Bandit problem with a set of N agents, is studied, which aims for a solution with which agents will never share their local observations with a central entity, and will be allowed to only share a private copy of his/her own information with their neighbors.
Differentially-Private Federated Linear Bandits
TLDR
This paper devise \textsc{FedUCB}, a multiagent private algorithm for both centralized and decentralized (peer-to-peer) federated learning, which provides competitive performance both in terms of pseudoregret bounds and empirical benchmark performance in various multi-agent settings.
Optimal Rates of (Locally) Differentially Private Heavy-tailed Multi-Armed Bandits
TLDR
An algorithm is proposed that can be seen as locally private and robust version of SE algorithm, which provably achieves (near) optimal rates for both instance-dependent and instance-independent regret.
Deep Learning for Privacy in Multimedia
TLDR
This tutorial offers theoretical explanations followed by examples with software developed by the presenters and distributed as open source of machine learning algorithms that provide users with more control on the multimedia information they share.

References

SHOWING 1-10 OF 34 REFERENCES
RAPPOR: Randomized Aggregatable Privacy-Preserving Ordinal Response
TLDR
This paper describes and motivates RAPPOR, details its differential-privacy and utility guarantees, discusses its practical deployment and properties in the face of different attack models, and gives results of its application to both synthetic and real-world data.
Corrupt Bandits for Preserving Local Privacy
TLDR
A frequentist algoritthm, KLUCB-CF, and a Bayesian algorithm, TS-CF are devised and a lower bound on the expected regret of any bandit algorithm in this corrupted setting is provided.
Differentially Private Contextual Linear Bandits
TLDR
This paper gives a general scheme converting the classic linear-UCB algorithm into a joint differentially private algorithm using the tree-based algorithm and gives the first lower bound on the additional regret any private algorithms for the MAB problem must incur.
Encode, Shuffle, Analyze Privacy Revisited: Formalizations and Empirical Evaluation
TLDR
This work revisits the ESA framework with a simple, abstract model of attackers as well as assumptions covering it and other proposed systems of anonymity, and examines the limitations of sketch-based encodings and ESA mechanisms such as data-dependent crowds.
Prochlo: Strong Privacy for Analytics in the Crowd
TLDR
A principled systems architecture---Encode, Shuffle, Analyze (ESA), which extends existing best-practice methods for sensitive-data analytics, by using cryptography and statistical techniques to make explicit how data is elided and reduced in precision, how only common-enough, anonymous data is analyzed, and how this is done for specific, permitted purposes.
On the Use of LSH for Privacy Preserving Personalization
TLDR
It is reported in this work how changing the nature of LSH inputs, which in this case corresponds to the user profile representations, impacts the performance of L SH-based clustering and the final quality of recommendations.
Differentially-Private "Draw and Discard" Machine Learning
TLDR
This work presents the mechanics of client and server components of "Draw and Discard" and demonstrates how the framework can be applied to learning Generalized Linear models and analyzes the privacy guarantees provided by the approach against several types of adversaries.
Guidelines for Implementing and Auditing Differentially Private Systems
TLDR
An initial set of "best practices" for developing differentially private platforms, techniques for unit testing that are specific to differential privacy, guidelines for checking if differential privacy is being applied correctly in an application, and recommendations for parameter settings are provided.
Crowd-Blending Privacy
TLDR
If a crowd-blending private mechanism is combined with a "pre-sampling" step, where the individuals in the database are randomly drawn from some underlying population as is often the case during data collection, then the combined mechanism satisfies not only differential privacy, but also the stronger notion of zero-knowledge privacy.
Differential Privacy for Multi-armed Bandits: What Is It and What Is Its Cost?
TLDR
This work derives and contrast lower bounds on the regret of bandit algorithms satisfying these definitions, and shows that for all of them, the learner's regret is increased by a multiplicative factor dependent on the privacy level $\epsilon$.
...
...