Dr. Hitesh Ballani of Microsoft Research

Abstract : Digital computing is approaching its fundamental limits just as compute-intensive workloads like machine learning are taking off. To address this, we are building a new kind of computer–an analog optical computer–to accelerate AI inference and hard optimization workloads. The computer has the potential to improve the efficiency and sustainability of these workloads by around 100x by stepping away from several fundamentally limiting aspects of general-purpose digital computing. It leverages chip-scale optical and electronic technologies from the consumer space that are low cost and scalable. In this talk, I will describe two generations of this computer that we have built, outline our roadmap for scaling, and discuss the importance of hardware-software co-design for such emerging computers and their potential for accelerating real-world problems in the post-Moore Law’s era.

Biography:

Bio: Hitesh Ballani is a partner researcher at Microsoft Research in Cambridge, UK. He leads the Future AI Infrastructure team, which incubates and innovates new technologies to sustain cloud and AI data centers, with a particular focus on new (optical) technologies for compute, storage, and networking, spanning novel components, devices, and full systems. Earlier, he worked on networked and storage system technologies that ship in Microsoft Windows. He received a PhD in Computer Science from Cornell University in 2009 and then joined Microsoft