Guest Talk by Dr. Niclas Kannengießer and Kevin Armbruster (TUM) on Tuesday, August 5, at 4pm

Title

Rethinking ML Development: A Sociotechnical Perspective

Abstract

Machine learning (ML) strongly influences societies by reshaping various professional and casual tasks. At the core of models stimulating such influences is ML development that has similarly crucial social impacts. Such impacts originate from how the choices are made about data to be used, collaboration, and ML workflow structure. In this talk, Niclas Kannengießer and Kevin Armbruster propose to rethink ML workflows through a sociotechnical lens. They will present their research on collaborative distributed machine learning, highlighting how decentralization could promote better alignment with European notions of digital sovereignty, while maintaining competitiveness in ML development. They will then look ahead to emerging directions in decentralized resource provisioning for ML as a foundation for future, open infrastructures. Turning a critical eye to the growing adoption of automated machine learning (AutoML), they will discuss opportunities and challenges for AutoML in decentralized settings and highlight open research avenues that need addressing for responsible, practitioner-centered automation. They will close by problematizing even centralized AutoML through ongoing research on human-in-the-loop ML, advocating for iterative, context-sensitive development processes that balance automation with practitioner goals and societal contexts. The talk aims to foster discussion on how ML can better align with social values, not only in use but in the way ML models are developed and iteratively improved.