Jay Yagnik, Head of Machine Perception Research at Google, presents the "Rapid Evolution and Future of Machine Perception" tutorial at the May 2017 Embedded Vision Summit.
With the advent of deep learning, our ability to build systems that derive insights from perceptual data has increased dramatically. Perceptual data dwarfs almost all other data sources in both its richness and its sheer size. This poses some unique challenges that have forced learning systems to evolve. This technical progress has enabled learning systems to be adopted in mainstream consumer products across the industry, such as Google Photos and YouTube, where learning systems have clearly proven their usefulness.
In this talk, Yagnik reviews the key ingredients of recent progress in machine perception. He also explores the substantial gaps that still need to be filled, and highlights some emerging applications that illustrate the potential future impact of this technology.