[Paper Review] Incentive Design for Efficient Federated Learning in Mobile Networks: A Contract Theory Approach

Jiawen Kang, Zehui Xiong+4 more|arXiv (Cornell University)|May 16, 2019

Privacy-Preserving Technologies in Data20 references34 citations

TL;DR

The paper uses contract theory to design an incentive mechanism that motivates mobile devices with high-quality data to participate in federated learning, addressing information asymmetry and resource costs. It shows improved learning efficiency and demonstrates feasibility via MNIST experiments.

ABSTRACT

To strengthen data privacy and security, federated learning as an emerging machine learning technique is proposed to enable large-scale nodes, e.g., mobile devices, to distributedly train and globally share models without revealing their local data. This technique can not only significantly improve privacy protection for mobile devices, but also ensure good performance of the trained results collectively. Currently, most the existing studies focus on optimizing federated learning algorithms to improve model training performance. However, incentive mechanisms to motivate the mobile devices to join model training have been largely overlooked. The mobile devices suffer from considerable overhead in terms of computation and communication during the federated model training process. Without well-designed incentive, self-interested mobile devices will be unwilling to join federated learning tasks, which hinders the adoption of federated learning. To bridge this gap, in this paper, we adopt the contract theory to design an effective incentive mechanism for simulating the mobile devices with high-quality (i.e., high-accuracy) data to participate in federated learning. Numerical results demonstrate that the proposed mechanism is efficient for federated learning with improved learning accuracy.

Motivation & Objective

• Motivate mobile devices with high-quality local data to join federated learning under information asymmetry.
• Model data quality as a contract type and design resource-reward bundles.
• Maximize the task publisher's profit while ensuring data owners' participation and truthfulness.

Proposed method

• Model federated learning as a monopoly with a task publisher and data owners.
• Define data quality as a contract type p_theta_n and formulate contract bundles (R_n(f_n), f_n).
• Impose Individual Rationality (IR) and Incentive Compatibility (IC) constraints and transform to Local Downward Incentive Constraints (LDIC).
• Derive optimal rewards R_n and CPU resources f_n by solving a concave optimization problem via CVX under time and budget constraints.
• Show that solving the relaxed problem followed by monotonicity enforcement yields feasible contracts and higher publisher profit than Stackelberg models.

Experimental results

Research questions

• RQ1 How can contract theory be applied to design incentives for data owners with heterogeneous data quality in federated learning?
• RQ2 What is the impact of data quality (type) and resource contributions on the optimal contracts and overall system profit?
• RQ3 Can IR and IC constraints be satisfied while maximizing the task publisher's profit under time and budget limits?

Key findings

• The contract-based incentive mechanism attracts higher-quality data and improves federated learning efficiency.
• IR and IC constraints are satisfied by the proposed contract, with data owners choosing only their designated contracts.
• The publisher's profit increases with more data owner types and is higher under the contract model than in Stackelberg game models.
• Increasing the upper bound on local data accuracy (higher-type prevalence) raises publisher profit.
• The proposed method achieves higher profit than symmetric-information Stackelberg benchmarks while maintaining data-owner utilities nonnegative.