Collaborative and Federated Data Analytics Beyond Predictive Modeling
The tremendous increase in computation capabilities of edge devices, along with the rapid market infiltration of powerful AI chips, has led to explosive interest in collaborative analytics, such as federated learning, that distribute model learning across diverse sources to process more of the user's data at the origin of creation. To date, these efforts have focused mainly on predictive modeling, where the goal is to create a global or personalized predictive map (often a deep network) that leverages knowledge from different sources while circumventing the need to share raw data. In this talk, I argue that predictive modeling, without untangling the nature of heterogeneity across users, can lead to swift and evident failures. With this in mind, I then present: i) A descriptive framework capable of extracting interpretable and identifiable features that describe what is shared and unique across diverse data datasets, ii) A prescriptive framework that utilizes the learned features for collaborative sequential design wherein dispersed users effectively distribute their trial & error efforts to improve and fast-track the optimal design process. I conclude the talk by describing some of our real-world prototyping and testing efforts.
Bio: Raed Al Kontar is an assistant professor in the Industrial & Operations Engineering department at the University of Michigan and an affiliate with the Michigan Institute for Data Science. Raed's research focuses on collaborative, distributed, and decentralized data science. Raed obtained an undergraduate degree in civil & environmental engineering and mathematics from the American University of Beirut in 2014 and a master's degree in statistics in 2017 and a Ph.D. degree in Industrial & System Engineering in 2018, both from the University of Wisconsin-Madison. Raed's research is currently supported by NSF, including a 2022 CAREER award, NIH, NLM, and various industry collaborators