Abstract: Machine Learning (ML) on devices is leading a paradigm shift in the world of machine learning. The driving force behind this is the goal to bring ML closer to user data to ensure privacy and a good user experience through sub-second latencies. We are witnessing innovations across the stack from hardware to applications; Number of frameworks like coreML, WinML, TFLite are emerging to solve the problem but, as an enterprise, trying to adopt a strategy around this is not easy. Applications running on variety of laptops and mobile devices need number of aspects to be considered while designing and architecting a solution to solve this at scale. In this session, we will cover some of these aspects to furnish real results and keep up with the evolving ecosystem.
The talk will be a combination of technical and business choices. For example, choosing between coreML, TFLite and WinML depends not only on the feature set t and the performance numbers they provide but also developer agility to experiment. Similarly, thinking about a solution that runs on GPU, or a combination of GPU and CPU is not just a technical decision but a business choices too because selecting a wide range of devices is needed for business reasons but increases the technical challenge significantly.
The biggest takeaway for the audience will be getting an understanding of technology stack of ML on devices. It will cover an overview of various technologies that come together for developingML on devices but also how to think about various aspects while designing a solution to ensure it is covering all the different aspects.
Bio: Coming Soon!