Currently, among the three elements of artificial intelligence, data is growing rapidly at the speed of a big explosion, and algorithms are also improving accordingly. The computing power, which has always been in a scarce state, is one of the major challenges in the development of today's AI. With the innovation and practice of the industry, various AI accelerators have been born. Among them, how to conduct unified and effective AI performance measurement and continuously improve AI performance on this benchmark has been a topic of concern to the entire AI practitioner for a long time.
htmlOn August 28, the AICC2019 Artificial Intelligence Computing Conference will be held in Beijing. Experts and celebrities from famous enterprises and institutions such as Alibaba , Cambrian , Baidu , SPEC, Intel , Kun Cloud Technology will gather at the first AI computing and benchmarking theme forum to share cutting-edge insights on AI chips, AI acceleration and AI benchmarking, and discuss the improvement of AI computing power.
AI computing and benchmark test: The key to the accelerated development of the AI industry
The scarcity and expensive computing power have always been a pain point for AI practitioners, especially algorithm researchers. Moore's Law, the famous chip industry, can no longer keep up with the development of AI's requirements for computing power.
Recently, the design of AI chips has begun to develop towards special chips. "Special cores and dedicated" can show geometric growth in computing power in specific fields, and can achieve dozens to hundreds of times improvements. The "2018-2019 China Artificial Intelligence Computing Power Development Evaluation Report" jointly released by IDC and Inspur believes that computing power is the core element and main driving force for the overall development and rapid application of artificial intelligence systems. Discussing the development direction of AI computing power and uniformly measuring the performance of various types of AI chips has become a topic that AI practitioners pay close attention to.
It is also understood that at the 2019 International Smart Computer Conference held not long ago, the International Testing Council (Bench Council) jointly with relevant institutes of the Chinese Academy of Sciences, the National Supercomputing Shenzhen Center and other scientific research institutions, issued five test standards, including "smart terminal (IoT AI), edge intelligence (Edge AI), smart data center, smart supercomputer (HPC AI), scientific big data". The release and later implementation of these artificial intelligence testing standards will also be expected to guide the entire AI industry to develop healthily and orderly under the "standards".
experts gather, and the advanced technologies and experience of AI computing power are all among them
It is reported that in response to the topics of AI chip development, AI performance evaluation standard system, and the latest AI acceleration platform that are widely concerned by the industry, 7 top experts and scholars from different industries and fields from around the world will share the latest R&D results and industry applications from laboratories and enterprises in this AICC 2019 Artificial Intelligence Computing Conference.
AI Computing and Benchmark Testing Forum Agenda (based on actual implementation)
Among them, Xu Lingjie, Director of Alibaba Group , Alibaba Cloud Intelligent Business Group, will bring us a speech on "Post-General Chip Era: AI Application Leads the Change of Data Centers". His view is very sharp - "AI applications that cannot be accelerated by special chips will be eliminated." In this speech, he will discuss how the trend of AI chip specialization will affect AI applications and promote the transformation of data centers.
Cambrian is a pioneer in the field of AI chips and one of the companies focusing on the development of dedicated computing chips. Cambrian Vice President Liu Daofu will share "Development and Application of Artificial Intelligence Chip Technology" at the forum.
Baidu paddlepaddle chief R&D architect Dong Daxiang, shared "Padpaddle Large-scale Distributed Training and Application". PaddlePaddle is a fully functional end-to-end open source deep learning platform, integrating deep learning training and prediction framework, model library, tool components and service platform. The super-large-scale parallel deep learning capability is one of the features of this platform.
SPEC has always been an authoritative performance evaluation agency and has a good reputation in benchmarking. SPEC ML Chair Kang Zhenjian will bring "SPEC and Machine Learning Benchmark Test", which discusses AI computing under different hardware and different computing frameworks, and can also have unified benchmarks and methods to test performance.
Intel machine learning expert and Mlperf benchmark expert Ma Guokai will bring "Introduction to MLPerf Machine Learning Benchmark Test".
Founder and CEO of Kunyun Technology and Executive Director of Kunyun Artificial Intelligence Application Innovation Research Institute will bring "High-performance AI acceleration, providing the next generation of artificial intelligence computing platform".
Intel Big Data Analysis and Artificial Intelligence Innovation Institute, Senior Project Manager Le Pengfei will bring "Unified End-to-End Big Data Analysis and Artificial Intelligence Platform".
VMware chief scientist He Jin will bring you the theme sharing of "vSphere - The Premiere Platform for AI and by AI".
(The picture in the article is authorized by Beijing Youfan Technology Co., Ltd. for use in China Daily)
Source: China Daily
