China Securities Network News (Reporter Peng Siyu) May 19, market research organization Frost Sullivan (referred to as "Sullivan") released the "China Deep Learning Software Framework Market Research report (2021)." According to the report, the results of a multi-dimensional analysis of manufacturers of deep learning software frameworks in the Chinese market show that Baidu, Inc. ranks first in terms of competitiveness and has significant advantages in terms of application ability, technical ability and ecological ability. Followed by Meta's PyTorch and Alphabet Inc-CL C's TensorFlow, Amazon.Com Inc's MXNet and Huawei's MindSpore ranked fourth and fifth respectively.
Sullivan believes that the competition pattern of the mainstream deep learning software framework is becoming clearer, and there are three key factors to win the Chinese deep learning software framework market, namely, framework performance, hardware support and market performance. it reflects the manufacturer's product ability, ecological ability and application ability respectively. Relevant manufacturers need to cross the four barriers of technology, talent, capital and brand, and compete fiercely around the three elements of success.
In terms of products, the report believes that with the change of products from mainstream manufacturers, the deep learning software framework has entered a stage of accelerated development. China is rapidly carrying out the systematic layout of open source development frameworks, with accelerated upgrading of self-developed open source deep learning software frameworks such as Baidu, Inc. Flying Propeller PaddlePaddle, Tencent YouTu NCNN, Huawei MindSpore, Ali XDL and so on.
In terms of the market, the report believes that the wider the application coverage and the larger the market share, the more it reflects that the enterprise has a stronger and more perfect ability to meet the customized needs of different customers. In addition, the more the number of application industries of the deep learning software framework and the more the number of subdivided application scenarios, it proves that the industry has a wide range of applications, indicating that enterprise products are widely used in all scenarios.
According to Sullivan statistics, there were 158 financing events involving more than 100 million dollars in AI in 2021, among which autopilot, smart health care and intelligent voice financing significantly exceeded other areas, becoming hot areas for deep learning software framework applications, and greater development opportunities will emerge.
On the ecological front, the report said that the focus of competition in the industry will shift from model base to ease of use and hardware adaptation optimization, and high-level language interface and hardware adaptation optimization will become the key to building barriers to open source frameworks. On the one hand, the high-level language interface encapsulates the key modeling, training and other functions in the back-end framework to reduce the threshold of research and development; on the other hand, hardware adaptation optimization attempts to solve the problems of complex adaptation and uneven performance caused by various hardware compilation tools. Unified compilation tools and compilation languages have become the layout focus of the mainstream open source development framework.
It is reported that Baidu, Inc. flying paddle PaddlePaddle is one of the earliest open frameworks launched in China. In 2021, Flying Propeller announced a total investment of 1.5 billion yuan in funds and resources to fully start the "Great Navigation" plan. Flying Propeller has set up "Baidu, Inc. Flying Propeller artificial Intelligence Industry enabling Center" in many places throughout the country, which has been widely used in intelligent manufacturing, intelligent finance, intelligent medical and other business scenes. By the end of 2021, the flying oar platform has gathered 4.06 million developers, created 476000 models and served 157000 enterprises and institutions. In terms of adaptation chips, in addition to the Kunlun core developed by Baidu, Inc., there are 22 domestic and foreign hardware manufacturers that adapt and optimize flying propellers, with a total of 31 chips, covering the mainstream chips in the market.