Sanjay Krishnan (UChicago)- Towards Efficient and Transparent Cross-Border Data Movement for Artificial Intelligence Applications

The next frontier of intelligent computing is to build AI systems whose decisions affect the real world (or at least go beyond a web browser). “Closing-the-loop” with AI requires solving a subtle challenge of cross-border data movement, where data may cross process, network, or trust boundaries. This talk describes my research over the last decade studying how data borders introduce data engineering bottlenecks that can preclude the design of performant, reliable, and trustworthy systems. While there is no silver bullet, two key solution themes emerge: (1) the importance of scalable logging and data provenance, (2) reliable approximate computing. I will present these themes in the context of a distributed AI-model serving platform called StreamServe which has been applied to problems ranging from network telemetry to robotic teleoperation. The solutions that my group has pioneered for (1) and (2) are deeply rooted in the rich algorithmic and theoretical traditions of the database research community, and I will emphasize how classical problem statements resurface in this new AI era.