1: \begin{abstract}

2: \emph{Bayesian optimization} (BO) is a prominent approach to optimizing expensive-to-evaluate black-box functions. The massive computational capability of edge devices such as mobile phones, coupled with privacy concerns, has led to a surging interest in \emph{federated learning} (FL) which focuses on collaborative training of \emph{deep neural networks} (DNNs) via \emph{first-order optimization} techniques.

3: However, some common machine learning tasks such as hyperparameter tuning of DNNs lack access to gradients and thus require \emph{zeroth-order/black-box optimization}.

4: This hints at the possibility of extending BO to the FL setting (FBO) for agents to collaborate in these black-box optimization tasks.

5: This paper presents \emph{federated Thompson sampling} (FTS) which overcomes a number of key challenges of FBO and FL in a principled way: 

6: We (a) use \emph{random Fourier features} to approximate the Gaussian process surrogate model used in BO, which naturally produces the parameters to be exchanged between agents,

7: (b) design FTS based on \emph{Thompson sampling}, which significantly reduces the number of parameters to be exchanged,

8: and (c) provide a theoretical convergence guarantee that is robust against heterogeneous agents, which is a major challenge in FL and FBO.

9: We empirically demonstrate the effectiveness of FTS in terms of communication efficiency, computational efficiency, and practical performance.

10: \end{abstract}

11: