Guided Research Milena Zahn

Towards Collaborative Business Models for

Interorganizational Federated Machine Learning

The demand for large datasets with good quality is a continuous challenge in training Machine Learning (ML) models to obtain desired outcomes. Overcoming the persistent bottleneck of insufficient training data is challenging. Moreover, data silos are held by a handful of entities whereas most organizations only have datasets that are insufficient for complex ML implementations. Hence, organizations need to collaborate to train powerful ML models. Federated Machine Learning (FedML) is a rising technology in the ML area that enables collaborative development of a consolidated model while keeping data decentralized. Contrary to the traditional centralized ML approach, data remains decentralized with models moving to the data. Only the resulting update gradients of the locally trained models are being aggregated centrally. Since the training data is kept locally, the issue of overcoming data silos is addressed. The implementation of this technology can be observed mainly within single organizations, however, FedML yields high potential for enabling cooperation beyond organizational boundaries. Overcoming data silos is not only a technical question, but also a political one. Organizations’ willingness to voluntarily share data is low for several reasons, including privacy- or trust-issues. In particular, interorganizational collaborations deal with interests of several stakeholders, resulting in the necessity of governance processes. There are existing concepts for governance of interorganizational collaborations and their business models. However, it remains unclear whether those can be applied to FedML specific processes. For this purpose, differences and commonalities with regard to FedML technology must be determined. This guided research investigates interorganizational FedML collaborations and works towards a business models by aiming to answer the following research questions:

• RQ1: What are governance-related challenges of interorganizational Federated Machine Learning in distinction to prevailing collaborative Business Models?
• RQ2: What are aspects of interorganizational Federated Machine Learning Business Models, and which attributes could be used to extend the Business Models Canvas by Osterwalder?
• RQ3: What are the foundations of governing Machine Learning processes, and how can they be derived to interorganizational Federated Machine Learning?

The research methodology is based on a literature review and semi-structured interviews. The literature review is conducted to identify governance-related challenges in FedML and to extract fundamentals for governing interorganizational collaborations. Based on existing literature, a proposal for a governance concept can be developed. This concept will be revised with the findings of semi-structured interviews, and new insights will be empirically collected and incorporated.