Topic |
Time |
Presenter |
AI & GAN in scientific research |
30 Min |
Daochen Shi |
BigData tools Rapids & Unsupervised learning |
45 Min |
Miles Xu |
GPU new technology update
|
45 Min |
Yi Cheng |
Discussion |
|
|
Daochen Shi is a deep learning solution architect of NVIDIA. He previously received the M.S. degree from Peking University on applied mathematics. His research interests include enhancing computer vision with deep learning models, working towards more-integrated systems and providing end-to-end solutions from edge devices to clusters.
Abstract: GPU enabled large scale of AI developing, especially in deep learning areas. This talk would go over the deep learning concept and discuss some state-of-the-art models on computer vision areas as well as recent works from NV research.
Xu Ming studied in Computer School of Wuhan University for 8 years for PhD. He have a lot experiences in high performance computing, especially in CUDA programming which he has begun to use since 2008. Before joined NVIDIA, He has accelerated a radar simulation software which is originally written in Matlab. The performance boost is above 200X. Since He joined NVIDIA, he has collaborated with Tsinghua NVAIL to accelerate their algorithms. For example, a topic model demo based on TensorFlow have got a 7.5X accelerating ratio.
Abstract: Use Rapids to accelerate BigData application on GPU, and one semi-supervised learning reference.
Yi Cheng: NVIDIA senior solution architecture , focusing on HPC over 7 years , experienced in HPC cluster system solutions and beyond , include GPU computing and DGX supercomputer , HPC and AI applications , CUDA and OpenACC programming skills .
Abstract : Turing architecture : NVIDIA’s new GPU architecture generation , bring new features for HPC and AI , provide more computing performance .
DGX-2 & DGX-POD : DGX-2 is a supercomputer with 2PFlops amazing computing performance , it has many innovations for GPU computing system , DGX-POD is a great GPU cluster solution , both hardware and software , it is convenient for you to build , extend and manage .