Deep Learning, Hardware, and TensorFlow 21/09/16 KB L.T. 1.5 14:00Published: Friday, 16 September 2016
Deep Learning, Hardware, and TensorFlow by Pete Warden (Google) on 21/09/16 KB L.T. 1.5 at 14:00
The rise of deep learning as a solution for many audio, image, NLP, and other ML problems and its computationally-intensive nature means that new approaches for hardware have become attractive. This seminar will discuss the work that Google has been doing with TensorFlow to support different platforms, with a focus on ARM and DSP solutions for mobile and embedded deployment of models. Topics will include eight-bit arithmetic, model minimization, and cascade approaches to power minimization.
Pete Warden is the technical lead of Google's TensorFlow deep learning framework. He was founder and CTO of Jetpac, acquired by Google in 2014 for its mobile deep learning technology, and previously worked at Apple on GPGPU optimizations after a career in the game industry. He's the author of several books on data for O'Reilly, and is a Manchester CS graduate.
Refreshments served after the seminar at 3pm in the Staff tearoom.