Seminar: Intelligent Silicon for AI at Scale
Building AI systems at scale is hard. As the prediction accuracy of learning systems improves with larger data and modules, the computational requirements grow non-linearly. This challenge is compounded with emergent trends in AI such as reinforcement learning, cloud to edge orchestration, and real-time decision making. While much of the AI revolution has been enabled by hardware, its forward-looking evolution is limited by the end of Moore's law.
This talk explores the role of intelligent silicon in enabling AI at scale. Case studies in image recognition and video analytics systems are used to highlight the infrastructure challenges of scaling-out real-life AI applications. Intelligent silicon using Micron's 3D Xpoint storage class memory is emerging as one of the key enablers of scale-out infrastructure. Key new features such as memory visualization, memory and storage convergence, and temp storage acceleration are anlayzed in context of their impact in the in-memory analytics use case, commonly found in today's recommendation engines.
About the speaker: Samir is an engineering leader with specialization in video analytics, real-time learning, and storage infrastructure. He currently leades Micron's 3DXP Storage Class Memory teams in building next-generation compute systems for new and emerging AI applications. Prior to joining Micron, Samir was the CEO and founder of SCUTI AI, a startup focused at the intersection of deep learning and scale-out infrastructure. Samir has a PhD in Mechanical Engineering from the Ohio State University specializing in signal processing and control systems.
Hosted by Prof. Cheena Srinivasan