- This event has passed.
OurCrowd Portfolio Company Featured: Quantization of Neural Networks at Hailo
May 7, 2020 @ 13:00 - 14:00 EDT
Mark Grobman, VP of Machine Learning & Applications at OurCrowd Portfolio company, Hailo will be leading a webinar on Quantization of Neural Networks at Hailo. Mark leads the ML activities at Hailo since its inception. Mark has multi-disciplinary experience in varied fields such as deep learning, high-speed digital systems and neuroscience. Mark is an alumni of the Intelligence Corps Technology Unit, and holds a double B.Sc. in Physics and Electrical Engineering from the Technion and an M.Sc. in Neuroscience from the Gonda Multidisciplinary Brain Research Center at Bar-Ilan University.
Quantization is a key component in the efficient deployment of deep neural networks. 8-bit quantization holds the promise of 4x reduction in model size and an x16 reduction in compute and power consumption but can result in severe penalty to the net’s performance. At its heart, quantization is a simple trade-off between dynamic range and precision. Finding the local optimum for each layer is rather simple but the complex way in which changes at the output of individual layers affect the output of the whole network is what makes successful quantization of neural networks tricky. One simple way to address this is by using greedy algorithms which attempt global optimization by iteratively applying local optimizations. While conceptually simple, these methods perform well and are cheap to implement.In this webinar, we give a brief overview of the principles behind neural network quantization, followed by a review of two techniques recently developed at Hailo: Equalization by inversely proportional factorization (presented at ICML2019) and bias-correction (presented at ECV workshop, CVPR2019). When used in combination, these methods enable fast post-training quantization to 8-bit while achieving state-of-the-art results. A Q&A session will follow the presentation.
Register for free here