Amod Jai Ganesh Anandkumar
Talk 1: The Intelligent Edge - A Convergence of Hardware, Software, and Machine Learning
Abstract: Edge devices are growing more powerful by the day and can be much more than just a pipe between sensors and the cloud. They can serve as independent decision making platforms capable of running analytics while keeping the data on device. On the other hand, deep learning models have grown larger, requiring more compute and data, making it unsuitable for resource-constrained edge devices without modifications. This talk will provide an introduction to techniques for developing and deploying efficient and performant deep learning models on the edge from the ML and systems perspectives. Specifically, we will look at edge-friendly neural network architectures, model compression techniques such as pruning, quantization, and knowledge distillation, and selecting and tuning inference runtimes.
Talk 2: Smart Devices On Wheels: Autonomous Electric Software-Defined Vehicles
Abstract: The automotive industry is undergoing a massive transformation with advances in autonomy, connectivity, and electrification. This comes with significant shifts in electrical/electronic architecture, middleware, and application software. Simultaneously, consumers expect their cars to be more like smartphones - offer regular updates, apps on demand, personalization, and a rich immersive multimedia experience. This talk will provide an overview of the evolution of the car as a hardware - software - services platform and highlight emerging trends in the automotive industry.
Talk 3: A Neuroscientist, a physicist, and an ML engineer meet in an Uber pool: Insights into inter-disciplinary industry research
Abstract: Real-world problems are complex and span multiple disciplines. Solving such problems requires systems thinking and inter-disciplinary expertise. In this talk, we will take Driver Monitoring Systems as an example and illustrate the challenges arising from the variability of the automotive environment, limitations on sensing and compute, and complexities of biosignal interactions with mental and physiological states. We will look at how to formulate a system-level solution leveraging expertise involving neuroscience for understanding mental states, physics for designing sensors, and deep learning for extracting subtle signals from massive amounts of data.
Qualifications: PhD, Loughborough University, 2011
Affiliation: HARMAN International, Bangalore
Position: Director of Research
Email: [email protected]