2024-08-05
한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina
The west wind of the house comes from Aofei Temple
Quantum Bit | Public Account QbitAI
The talent black hole effect of the big model is still continuing.
Quantum位 learned that the latest heavyweight to appear on the transfer list of big model experts is a 90s AI expertZhang Xiangyu, one of the four authors of ResNet, Sun Jian’s first doctoral student in deep learning, and winner of the Future Science Award.
Now, he appears in the domestic large-scale model unicornStep StarThere are rumors of signings.
Moreover, it is said that there is more than one AI expert named Jieyuexingchen, one of the six little strong ones, recently.
However, as of press time, Jieyuexingchen has not yet made a clear response.
△Image source: Future Technology Awards official website
In 2008, Zhang Xiangyu was admittedXi'an Jiaotong University; One year later, I applied to change my major to software engineering.
In 2011, Zhang Xiangyu won the Finalist Award in the American Mathematical Contest in Modeling (MCM), setting the best result in the history of Xi'an Jiaotong University's participation in the competition at that time.
In 2012, Zhang Xiangyu received a bachelor's degree in software engineering from the School of Software, Xi'an Jiaotong University.
Afterwards, Zhang Xiangyu was selected for the Xi'an Jiaotong University andMicrosoft Research AsiaJoint doctoral training program,Studied under Sun Jian and He KaimingIn 2017, he obtained his Ph.D.
It was during their time at Microsoft Research Asia that He Kaiming, Sun Jian, Ren Shaoqing and Zhang Xiangyu jointly completed the famousResNet, Zhang Xiangyu is responsible for the underlying framework and coding.
ResNet, or residual learning, is a research topic in the field of computer vision, but its core idea, residual connection, has become a basic component of modern deep learning models.
AlphaGo Zero, which started the last AI boom, was created by combining ResNet+reinforcement learning+Monte Carlo search.
ChatGPT, which started the latest AI craze, also uses residual connections in the "T" of the Transformer network.
The original ResNet paper won the CVPR Best Paper Award in 2016.
Last year, He Kaiming, Sun Jian, Ren Shaoqing and Zhang Xiangyu won the 2023 Future Science Award for this research.Mathematics and Computer Science Award”, which shows its far-reaching impact.
Currently, this paper has been cited more than 230,000 times on Google Scholar, making it one of the most popular deep learning papers of this century:
In July 2016, Sun Jian joined Megvii as chief scientist and head of Megvii Research Institute. After graduating with a Ph.D., Zhang Xiangyu followed his mentor and joined theWide-eyed。
Zhang Xiangyu's research directions include high-performance convolutional network design, AutoML and automated neural network architecture search, and pruning and acceleration of deep models.
In 2017 and 2018, Zhang Xiangyu completed another masterpiece - Efficient Convolutional Neural Network for Mobile DevicesShuffleNetTwo parts.
Co-founder of Dark Side of the Moon and head of the algorithm teamZhou Xinyu, which is also a joint work.
ShuffleNet later influenced various mobile phone millisecond-level face unlocking technologies, including Apple's 3D face unlocking.
In 2020, Zhang Xiangyu was ranked 4th on the list of "AI 2000 Most Influential Scholars in Computer Vision in the World".
In addition, Zhang Xiangyu has done many other research projects. Google Scholar data shows that his papers have been cited 320,510 times and have an h-index of 59.
In 2024, the flow of AI talents has not decreased, but their choice is not all to start a business.
People in the industry often have strong personal experience with AI technology. Many of them, like Zhang Xiangyu, choose to join existing companies and surf the new trend passionately.
And Step Star has quietly recruited many talented people who have made this choice.
Quantum位 learned that Tencent’s Yu Gang and MSRA’s Duan Nan have also joined the Step Star team.
Yu Gang, current CEO of Step StarPrincipal Researcher.
Prior to this, he served as research director at Tencent; before that, he was the team leader of Megvii Detection.
Yu Gang graduated from Shandong University with a bachelor's degree in 2007 and received a master's degree in computer science from Shanghai Jiao Tong University in 2010.
He then went to Nanyang Technological University to pursue a doctorate degree under the tutelage of Yuan Junsong. During this period, he interned at Microsoft Research and joined Megvii after graduating in 2014.
Mainly responsible for the research on detection, segmentation, tracking, skeleton, motion behavior, etc. as well as the implementation of algorithms.
In 2017, Yu Gang led the team to participate in the 2017 COCO+Places Challenge, and won the first place in detection and the first place in human pose estimation. In 2018, he led the team to participate in the COCO+Mapillary Challenge again and won four first places.
During his five years at Megvii, Yu Gang and Sun Jian, Megvii's chief scientist and director of Megvii Research Institute, co-authored ten papers.
At the end of 2019, Yu Gang confirmed his resignation on Zhihu and announced that he would join Tencent PCG Light and Shadow Research Lab to continue his research on CV-related algorithms and technologies.
This year, Yu Gang had 2 papers accepted by ECCV2024 and 2 papers accepted by CVPR; at the same time, he also served as the regional chairman of CVPR.
Currently, Yu Gang’s Google Scholar citations are 19,000.
NextMSRA Duan Nan——In fact, in May this year, Quantum位 shared with everyone the news that he joined Step Star.
Before leaving, Duan Nan was a senior principal researcher at MSRA and research manager of the natural language computing team.
He is mainly engaged in research on natural language processing, multimodal intelligence, code intelligence, machine reasoning, etc. He has served as program chair and field chair of NLP/AI academic conferences many times.
He is also a part-time doctoral supervisor at the University of Science and Technology of China and Xi'an Jiaotong University, and a part-time professor at Tianjin University.
Duan Nan joined MSRA after graduating with a Ph.D. from Tianjin University. Zhou Ming, one of the most accomplished Chinese in the field of NLP today and former vice president of MSRA, is his mentor.
He worked at MSRA for 17 years and 9 months.
During his work, many of Duan Nan's research results were successfully applied to Microsoft AI products, including Bing Search, Bing Ads, Microsoft News, Cortana voice assistant and Microsoft Xiaoice.
His research also includes Microsoft NüWA, a multimodal pre-training model jointly created by MSRA and Peking University, which directly achieved 8 SOTA awards when it was first released.
It is understood that Duan Nan is responsible for serving as the research leader and mentor of the NüWA series model family project.
At present, Duan Nan’s Google Scholar citations are approaching 19,000.
Other statistics show that Duan Nan and Jiang Daxin, former global vice president of Microsoft and founder and CEO of Step Star, have co-authored 44 papers.
In just a few months, Zhang Xiangyu, Yu Gang and Duan Nan were recruited, and the news of more AI talents joining was kept secret.
The step star must be very capable.
The reason is that first of allWell-recognized status in the industry。
Jieyuexingchen is already recognized as one of the "six strongest" domestic large-scale model startups:
Together with Dark Side of the Moon, Minimax, Zhipu Huazhang, Zero One Everything, and Baichuan Intelligence, we have been deeply engaged in the field of basic models. Due to the comprehensive evaluation of talents, technology, products, and market recognition, our valuation has quickly jumped to more than 1 billion US dollars.
In June, it was revealed that Jieyuexingchen's value in the investment and financing market had reached US$2 billion.
ThenA bastion of talent that attracts people as soon as they appear。
Founder and CEOJiang Daxin, former global vice president of Microsoft.
Without mentioning anything else, just looking at his final title at Microsoft, one can understand why many AI experts who have close ties with MSRA are rushing towards the embrace of the Leap Star (a dog head).
In 2011, Jiang Daxin started working at Microsoft Asia Internet Engineering Academy (STCA). In March last year, he was promoted to vice president of Microsoft and concurrently served as general manager of WebXT S+D (web experience, search and distribution) group of Asia Software Technology Center.
Then, when his career reached a new peak, he still started his own business and achieved the status quo he has today.
Data Manager of Step StarJiao Binxing, is to package yourself together with Jiang Daxin and join the new wave of AI entrepreneurship from Microsoft.
The company's system managerZhu Yibo, and also had work experience at MSRA.
More importantly, he is one of the few practitioners in China who has practical experience in system construction and management with more than 10,000 cards in a single cluster.
And the indispensable snowball-building tool - Step's self-developed models and products.
On the model side, its Step series of general large models, as well as its two toC products, Yuewen and Maopaoya, have been gradually put on the market for testing and are gaining greater recognition.
The above may explain why Leap Star has become the "AI talent black hole" we see.
The attraction of Step Stars to talents is the epitome of why the six little strong men of the big model are frantically “collecting talents”.
The flow of talent will continue.
References:
[1]https://se.xjtu.edu.cn/info/1044/2628.htm
[2]https://www.skicyyu.org/
[3]https://scholar.google.com.sg/citations?user=BJdigYsAAAAJ&hl=en
[4]https://nanduan.github.io/
Collect and organize the latest news information and make it available for free for everyone to read
