Advances and Open Problems in Federated Learning

Abstract

Federated learning (FL) is a machine learning setting where many clients (e.g. mobile devices or whole organizations) collaboratively train a model under the orchestration of a central server (e.g. service provider), while keeping the training data decentralized. FL embodies the principles of focused data collection and minimization, and can mitigate many of the systemic privacy risks and costs resulting from traditional, centralized machine learning and data science approaches. Motivated by the explosive growth in FL research, this paper discusses recent advances and presents an extensive collection of open problems and challenges.

Design and Analysis of Uplink and Downlink Communications for Federated Learning

“…A comprehensive review of these challenges can be found in [6]. In particular, despite being recognized as one of the primary bottlenecks of FL [2], [6], [7], research on the communication aspect in the FL pipeline has not been on par with the overview. There are also recent studies that focus on the communication system design [8]- [10], but they are either system-specific (e.g., cellular networks) or with high complexity beyond the current implementation capability (e.g., very high dimensional vector quantization).…”

Open-Source Federated Learning Frameworks for IoT: A Comparative Review and Analysis

“…The FL process typically includes the following steps [ 9 ]: Identifying a problem to be solved; Modifying the client’s application (optional); Simulating prototyping (optional); Training the federated model; Evaluating the federated model; Deploying FL at the server and clients. …”

“…Based on how data are distributed, FL systems can typically be categorized as horizontal or vertical FL systems [ 9 , 10 , 24 ].…”

“…There are two different types of FL systems depending on the scale of federation [ 9 , 10 ] ( Figure 3 ): Cross-silo systems have low scalable federation. They include organizations or data centers.…”

“…Currently, there is a number of surveys and reviews devoted to different aspects of implementing FL. For example, in [ 9 , 10 ], the authors present a detailed analysis of FL concepts and the architecture of FL systems, as well as privacy-aware issues, such as approaches to preserving data subject privacy and protecting against different types of attacks. Shokri et al [ 11 ] investigated basic membership inference attacks that assume detection for a data record and black-box access to a model if this record was in the model’s training data set.…”