Cohort-based Federated Learning Services for Industrial Collaboration on the Edge

Abstract

<div>Machine Learning (ML) is increasingly applied in industrial manufacturing, but often performance is limited due to insufficient training data. While ML models can benefit from collaboration, due to privacy concerns, individual manufacturers cannot share data directly. Federated Learning (FL) enables collaborative training of ML models without revealing raw data. However, current FL approaches fail to take the characteristics and requirements of industrial clients into account. In this work, we propose a FL system consisting of a process description and a software architecture to provide \acrfull{flaas} to industrial clients deployed to edge devices. Our approach deals with skewed data by organizing clients into cohorts with similar data distributions. We evaluated the system on two industrial datasets. We show how the FLaaS approach provides FL to client processes by considering their requests submitted to the Industrial Federated Learning (IFL) Services API. Experiments on both industrial datasets and different FL algorithms show that the proposed cohort building can increase the ML model performance notably.</div>

Evaluation of Hyperparameter-Optimization Approaches in an Industrial Federated Learning System

“…Industrial assets have access to a wealth of data suitable for machine learning models, however, the data on an individual asset is typically limited and private in nature. In addition to sharing the model within the company, it can also be shared with an external industry partner [1]. FL leaves possibly critical business information distributed on the individual client (or within the company).…”

“…In an industrial context, we expect to find heterogeneous clients due to varying environmental and operational conditions on different assets. Therefore, Hiessl et al [1] introduced a modified approach of FL in an industrial context and termed it Industrial Federated Learning (IFL). IFL does not allow arbitrary knowledge exchange between clients.…”

“…Instead, the knowledge exchange only takes place between clients that have sufficiently similar data. Hiessl et al [1] refer to this set of clients as a cohort. We expect the federated learning approach in a cohort to approximate the corresponding central learning approach.…”

“…The aggregation is coordinated by a server. Industrial Federated Learning (IFL) is a modified approach of FL in an industrial context [1]. In an FL setting, hyperparameter optimization poses new challenges and is a major open research area [2].…”

“…The optimization problem in FL is therefore referred to as federated optimization emphasizing the difference to distributed optimization [5]. In an IFL setting, additional challenges regarding industrial aspects arise [1]. In this section, we want to formulate the federated optimization problem and discuss the algorithmic challenges of FL in general, and in particular of IFL.…”