# Differentially-Private Federated Linear Bandits

@article{Dubey2020DifferentiallyPrivateFL, title={Differentially-Private Federated Linear Bandits}, author={Abhimanyu Dubey and Alex 'Sandy' Pentland}, journal={ArXiv}, year={2020}, volume={abs/2010.11425} }

The rapid proliferation of decentralized learning systems mandates the need for differentially-private cooperative learning. In this paper, we study this in context of the contextual linear bandit: we consider a collection of agents cooperating to solve a common contextual bandit, while ensuring that their communication remains private. For this problem, we devise \textsc{FedUCB}, a multiagent private algorithm for both centralized and decentralized (peer-to-peer) federated learning. We provide… Expand

#### Supplemental Presentations

Presentation Slides

#### Figures from this paper

#### 20 Citations

Federated Linear Contextual Bandits

- Computer Science, Mathematics
- ArXiv
- 2021

A collaborative algorithm called Fed-PE is proposed to cope with the heterogeneity across clients without exchanging local feature vectors or raw data, and achieves near-optimal regrets for both disjoint and shared parameter cases with logarithmic communication costs. Expand

Federated Multi-armed Bandits with Personalization

- Computer Science, Mathematics
- AISTATS
- 2021

Theoretical analysis proves that the Personalized Federated Upper Confidence Bound (PF-UCB) algorithm achieves an O(log(T )) regret regardless of the degree of personalization, and has a similar instance dependency as the lower bound. Expand

Exploiting Heterogeneity in Robust Federated Best-Arm Identification

- Computer Science, Engineering
- ArXiv
- 2021

This work proposes Fed-SEL, a simple communication-efficient algorithm that builds on successive elimination techniques and involves local sampling steps at the clients and introduces a notion of arm-heterogeneity that captures the level of dissimilarity between distributions of arms corresponding to different clients. Expand

Differentially Private Multi-Armed Bandits in the Shuffle Model

- Computer Science
- ArXiv
- 2021

This work addresses the multi-armed bandit problem, which is applicable to various areas such as recommender systems, online advertising and clinical trials, and embodies the well known exploration-exploitation trade-off between learning the environment and acting optimally based on the authors' current knowledge about the environment. Expand

Robust Federated Best-Arm Identification in Multi-Armed Bandits

- 2021

We study a federated variant of the best-arm identification problem in stochastic multiarmed bandits: a set of clients, each of whom can sample only a subset of the arms, collaborate via a server to… Expand

Cooperative Stochastic Bandits with Asynchronous Agents and Constrained Feedback

- 2021

Motivated by the scenario of large-scale learning in distributed systems, this paper studies a scenario where M agents cooperate together to solve the same instance of a K-armed stochastic bandit… Expand

Optimal Rates of (Locally) Differentially Private Heavy-tailed Multi-Armed Bandits

- Computer Science
- ArXiv
- 2021

This paper proposes an algorithm which could be seen as locally private and robust version of the SE algorithm, and shows it could achieve (near) optimal rates for both instance-dependent and instance-independent regrets. Expand

Differentially Private Federated Bayesian Optimization with Distributed Exploration

- Computer Science
- ArXiv
- 2021

The resulting differentially private FTS with DE (DP-FTS-DE) algorithm is endowed with theoretical guarantees for both the privacy and utility and is amenable to interesting theoretical insights about the privacy-utility trade-off. Expand

Local Differential Privacy for Regret Minimization in Reinforcement Learning

- Computer Science
- 2020

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. Expand

Decentralized Multi-Armed Bandit Can Outperform Classic Upper Confidence Bound

- Computer Science, Engineering
- ArXiv
- 2021

A fully decentralized multi-armed bandit algorithm is proposed for each agent, which twists the classic consensus algorithm and upper confidence bound (UCB) algorithm and guarantees each agent to achieve a better logarithmic asymptotic regret than the classic UCB. Expand

#### References

SHOWING 1-10 OF 57 REFERENCES

Differentially private, multi-agent multi-armed bandits

- Computer Science
- EWRL 2015
- 2015

Two algorithms built upon decentralized Time Division Fair Sharing method and upper confidence bounds are derived, where all decisions are taken based on private statistics, that provide regret guarantees that are almost as good as the non-private, multi-agent algorithm and demonstrate them empirically. Expand

Private and Byzantine-Proof Cooperative Decision-Making

- Computer Science
- AAMAS
- 2020

This work provides upper-confidence bound algorithms that obtain optimal regret while being differentially-private and tolerant to byzantine agents, and requires no information about the network of connectivity between agents, making them scalable to large dynamic systems. Expand

Robust Algorithms for Multiagent Bandits with Heavy Tails

- 2020

We study the heavy-tailed stochastic bandit problem in the cooperative multiagent setting, where a group of agents interact with a common bandit problem, while communicating on a network with delays.… Expand

Cooperative Multi-Agent Bandits with Heavy Tails

- Computer Science, Mathematics
- ICML
- 2020

This work proposes a decentralized multi-agent algorithm for the cooperative stochastic bandit that incorporates robust estimation with a message-passing protocol and proves optimal regret bounds for \textsc{MP-UCB} for several problem settings, and also demonstrates its superiority to existing methods. Expand

Distributed cooperative decision-making in multiarmed bandits: Frequentist and Bayesian algorithms

- Computer Science, Mathematics
- 2016 IEEE 55th Conference on Decision and Control (CDC)
- 2016

This work rigorously characterize the influence of the communication graph structure on the decision-making performance of the group and proves the performance of state-of-the-art frequentist and Bayesian algorithms for cooperative distributed algorithms for multi-agent MAB problems in which agents communicate according to a fixed network graph. Expand

Kernel Methods for Cooperative Contextual Bandits

- 2020

Cooperative multi-agent decision making involves a group of agents cooperatively solving learning problems while communicating over a network with delays. In this paper, we consider the kernelised… Expand

Differentially Private Contextual Linear Bandits

- Computer Science, Mathematics
- NeurIPS
- 2018

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. Expand

Decentralized Cooperative Stochastic Multi-armed Bandits

- Computer Science, Mathematics
- ArXiv
- 2018

This work designs a fully decentralized algorithm that uses a running consensus procedure to compute, with some delay, accurate estimations of the average of rewards obtained by all the agents for each arm, and then uses an upper confidence bound algorithm that accounts for the delay and error of the estimations. Expand

Kernel Methods for Cooperative Multi-Agent Contextual Bandits

- Computer Science, Mathematics
- ICML
- 2020

This paper considers the kernelised contextual bandit problem, where the reward obtained by an agent is an arbitrary linear function of the contexts' images in the related reproducing kernel Hilbert space (RKHS), and proposes an algorithm that provides near-optimal bounds on the per-agent regret. Expand

Differentially Private Federated Learning: A Client Level Perspective

- Computer Science, Mathematics
- ArXiv
- 2017

The aim is to hide clients' contributions during training, balancing the trade-off between privacy loss and model performance, and empirical studies suggest that given a sufficiently large number of participating clients, this procedure can maintain client-level differential privacy at only a minor cost in model performance. Expand