Upcoming Seminars

Title: Data-Driven Management for Autonomous Systems
Seminar: N/A
Speaker: Christopher Stewart, Ohio State University
Contact: Dorian Arnold, dorian.arnold@emory.edu
Date: 2019-10-18 at 10:30AM
Venue: MSC W201
  Download Flyer  Add to Calendar
Abstract:
Computer systems are increasingly autonomous; they sense their surroundings, process their operating context and take actions to achieve their goals. Powered by AI, autonomous systems are proliferating in health care, agriculture, sustainability and cloud computing. Examples include autonomous vehicles, automated medical diagnosis and run-scripts for large, complex distributed systems. Autonomous systems simplify programming, reduce costly human labor and improve energy efficiency. However, autonomous systems are inherently closed loop, making their compute needs hard to model. This talk describes the overarching approach taken by ReRout Lab @OSU to manage resources for autonomous systems. We advance the notion of data-driven management that excels when domain-specific training data is married with contextual, first-principles knowledge. We have used this approach to build state of the art self-flying systems and self-managing cloud systems. Bio: https://web.cse.ohio-state.edu/~stewart.962/bio.html
Title: Machine Learning for the Real World: Provably Robust Extrapolation
Seminar: N/A
Speaker: Anqi Liu, California Institute of Technology
Contact: Dorian Arnold, dorian.arnold@emory.edu
Date: 2019-10-25 at 10:30AM
Venue: MSC W201
  Download Flyer  Add to Calendar
Abstract:
The unprecedented prediction accuracy of modern machine learning beckons for its application in a wide range of real-world applications, including autonomous robots, medical decision making, scientific experiment design, and many others. A key challenge in such real-world applications is that the test cases are not well represented by the pre-collected training data. To properly leverage learning in such domains, especially safety-critical ones, we must go beyond the conventional learning paradigm of maximizing average prediction accuracy with generalization guarantees that rely on strong distributional relationships between training and test examples. In this talk, I will describe a robust learning framework that offers rigorous extrapolation guarantees under data distribution shift. This framework yields appropriately conservative yet still accurate predictions to guide real-world decision-making and is easily integrated with modern deep learning. I will showcase the practicality of this framework in an application on agile robotic control. I will conclude with a survey of other applications as well as directions for future work.