O'Reilly、Cloudera 主办
Make Data Work

用于深度学习的异步计算(Heterogeneous computing for deep learning)

此演讲使用中文 (This will be presented in Chinese)

Ming Yang (Horizon Robotics)
09:45–10:05 Saturday, 2017-07-15
地点: 紫金大厅A(Grand Hall A)
平均得分:: *****
(5.00, 1 次得分)

深度神经网络的再次复兴带来了许多应用领域里的性能的显著提升,特别是对图像识别和视频内容分析。 密集的计算和内存访问模式需要创新的异步计算,或者换句话说,需要使用专用协处理器来加速神经网络的训练或推理。本讲座将简要介绍深度神经网络计算,异步计算的概述和正在相互竞争的异步计算选项(如DSP、GPU、TPU、FPGA和ASIC)之间的比较。

The resurgence of deep neural networks has led to a great performance boost in many application fields, especially for image recognition and video content analysis. The intensive computation and memory access pattern demands innovative heterogeneous computing, or to put it another way, dedicated coprocessors to accelerate the neural network training or inference. Ming Yang offers a brief introduction to deep neural network computation as well as an overview and comparison of the competing heterogeneous computing options, such as DSP, GPU, TPU, FPGA, and ASIC.

Photo of Ming Yang

Ming Yang

Horizon Robotics

Ming Yang is the cofounder and vice president of software at Horizon Robotics. Previously, he was one of the founding members of the Facebook Artificial Intelligence Research (FAIR) team and a former senior researcher at NEC Labs America. Ming is a well-recognized researcher in computer vision and machine learning. His research interests include object tracking, face recognition, massive image retrieval, and multimedia content analysis. He holds 14 US patents and has over 20 publications in top conferences like CVPR and ICCV and 8 publications in top international journal T-PAMI, with more than 5,000 citations. During his tenure at Facebook, Ming led the deep learning research project Deep Face, which had a significant impact in the deep learning research community and was widely reported by media including Science magazine, MIT Tech Review, and Forbes. He has served as a member of the program committee for multiple top international conferences, including CVPR, ICCV, NIPS, and ACMMM. He has also been a reviewer for several top international journals, including T-PAMI, IJCV, and T-IP. As the leader of the NEC-UIUC team, Ming and his team achieved the best result in the TRECVid 2008 and 2009 Event Detection Evaluation. He was also a member of the NEC team that won first place in the ImageNet 2010 Large Scale Visual Recognition Challenge. He holds a BEng and MEng from the Department of Electrical Engineering at Tsinghua University and a PhD from the Department of Electrical Engineering and Computer Science at Northwestern University.



WeChat QRcode


Stay Connected Image 1
Stay Connected Image 3
Stay Connected Image 2


ORB Data Site