Quantization of Neural Networks – High Accuracy at Low Precision

A webinar by Hailo: Quantization of Neural Networks– High Accuracy at Low Precision, held by Hailo's VP Machine Learning & Applications, Mark Grobman. Quantization is a key component of efficient deployment of deep neural networks. 8-bit quantization holds the promise of 4x reduction in model size and an x16 reduction in both compute and power consumption but can result in severe accuracy degradation. At its heart, quantization is a trade-off between dynamic range and precision. Finding the local optimum for each layer is simple, however, the complex way in which the output of individual layers affect the output of the network is what makes quantization of neural networks tricky. One simple way to address this is by using algorithms which perform greedy global optimization by iteratively applying local optimizations. While conceptually simple, these methods have excellent performance 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. For more information and resources visit our website at https://www.hailo.ai/resources/ About Hailo: Hailo is a leading AI chipmaker whose mission is to enable smart edge technologies to reach their full potential. Hailo’s industry-leading Hailo-8™ edge AI processor enables datacenter-class performance on cameras and other edge devices. It features up to 26 tera-operations per second (TOPS). One small Hailo-8™ chip can process, in real-time, high-resolution videos, or multiple video streams and neural network models simultaneously. The chip has a very low power consumption, which means it needs less cooling and is more cost-effective and sustainable. We offer a range of standard form-factor modules which enable easy integration with edge devices, as well as a robust and mature software toolchain. Hailo’s novel, patented, dataflow architecture is finding various applications across several industries, including automotive, surveillance, smart city, industry 4.0, smart retail, and more. Connect with Hailo: Visit the Hailo WEBSITE: https://hailo.ai Join Hailo on LINKEDIN: https://www.linkedin.com/company/hail... Like Hailo on FACEBOOK: https://www.facebook.com/HailoTech Follow Mobileye on TWITTER: https://twitter.com/hailo_ai #neuralnetworks #Hailo #edge #machinelearning