"On April 9, 2020, Wang Endong, academician of Chinese engineering and chief scientist of Inspur Group, said in a speech on" smart computing source driven new infrastructure "at IPF 2020 (the abbreviation of Inspur cloud data center Partner Conference in 2020), that the smart computing center is the most important computing power production center and supply center in the smart era. And pointed out that:
"Computing power is productivity. Intelligent computing has transformed and upgraded the three elements of productivity, and finally driven the transformation and upgrading of human society. Intelligent computing turns workers from people to people. With artificial intelligence, workers can show exponential growth; Smart computing has turned data into a new means of production, from tangible to intangible, and it is used more and more; Intelligent computing has turned computing driven information equipment into a production tool, which is also an exponential growth, and the productivity has been unprecedented liberated.
We can see that AI centered technology and its ecology have become the decisive factor in the development of new forms of productivity, and will help the deepening of new infrastructure and anti epidemic.
On the same day, Li Hui, general manager of Inspur storage product line, said that "driven by artificial intelligence, big data and 5g new technologies, data has changed in essence. In the transition from 'manual collection and manual intervention' in the past to the new era of 'machine generation and machine processing', data should not only be stored and managed well, but also used well. The reflected Inspur storage concept is: cloud storage and intelligent use, Operational research new data ".
In the field of distributed storage, as13000-h of Inspur storage can be used in HPC storage scenarios including AI, which can meet the needs of AI training, that is, the high OPS required for massive small file concurrency. As13000-h single node can support the performance of tens of thousands of OPS.
Li Hui pointed out: "Inspur storage is oriented to the two platforms of distributed storage and centralized storage, and upgrades seven extreme capabilities - in addition to realizing extreme speed performance based on iturbo intelligent engine, Inspur storage also upgrades the extreme security of automatic risk perception, the extreme stability of six protection, the extreme capacity of EB level optimal utilization, the extreme cloudization based on iscmi cloud docking technology Integrate the extremely easy management of aiops intelligent operation and maintenance, so as to bring a safe, reliable, cost-effective, easy-to-use and easy-to-manage storage platform for the Intelligent Computing Center ".
In the field of centralized storage, Inspur storage is in a global basic storage performance test such as spc-1. The performance of 16 control storage of as5600g2 ranks first in the world, with a delay of less than 0.5 ms under a load of 7.52 million IOPs. This shows that as5000 series storage can well meet the performance requirements of AI reasoning for mixed read-write and ultra-low delay.
Here are some parts of the 20000 word long article CB insights: 2019 artificial intelligence trend translated by Langchao Hou Yanlu to reveal the exploration and practice of the concept of Intelligent Computing Center in the field of AI algorithm and application.
2019 AI trends
First, we evaluate each trend using the CB insights nextt framework.
The nextt framework tells enterprises to understand emerging trends and guide their decisions according to their risk tolerance.
Nextt uses data driven signals from concept to maturity to widespread adoption to assess trends in technology, products, and business models.
Meaning of x-axis and Y-axis of nextt frame:
Industry adoption (Y-axis): including the development momentum of start-ups in this field, media attention and customer adoption rate (partnership, customers and licensed transactions).
Market strength (x-axis): including market scale forecast, quality and quantity of investors and capital, R & D investment, profit record analysis, competition intensity, current transactions (M & A, strategic investment).
The area covered by x-axis and y-axis is divided into four quadrants:
namely:
1. Necessary (upper right corner of coordinates, necessary)
Industry and customer implementation / Adoption, as well as market and application understanding are emerging trends. For these trends, incumbents should formulate clear and clear strategies and measures.
2. Experimental coordinates (lower left corner, necessary)
There are few concepts or early trends of functional products, which have not been widely adopted. The experimental trend has stimulated the interest and proof of concept of early media.
3. Threatening coordinates (lower left corner, threatening)
Large predictable market forecasts and significant investment activities. This trend has been accepted by early adopters and may soon be adopted by a wide range of industries or customers.
4. Transitional (upper left corner, transition)
Trends are being adopted, but market opportunities are uncertain. As temporary trends become more widely understood, they may reveal more opportunities and markets.
We put the 25 trends found into the nextt box and get the following overview.
According to the four quadrants of the trend, there are 25 in total. Including: 6 necessary; 12 experiments; 4 threats; Three of the transition. These 25 trends are divided by scene: Computer Vision: 7; Natural speech processing / synthesis: 5; Intelligent prediction: 6; Architecture: 4; Infrastructure: 3.
1、 Necessary
1. Open source framework benefits from open source software, and the barrier to artificial intelligence is lower than ever before.
Google opened the tensor flow machine learning library in 2015.
The open source framework for AI works in both directions: it enables everyone to use AI. In turn, companies like Google benefit from a community of contributors that help accelerate their AI research.
Hundreds of users contribute to tensorflow on GitHub every month (a software development platform in which users can collaborate). Here are some companies that use tensor flow, from Coca Cola to eBay to airbnb.
After cooperating with researchers from NVIDIA, Qualcomm, Intel, Microsoft and other companies to create a lightweight modular deep learning framework, Facebook released caffe2 in 2017. The framework can be extended beyond the cloud and can also be applied to mobile applications. Facebook also ran pytorch, an open source machine learning platform for Python. On May 18, 2018, Facebook combined the two into one, "combined the beneficial features of caffe2 and pytorch into one package, and realized a smooth transition from rapid prototyping to rapid execution".
The number of GitHub contributors to pytorch has increased in recent months.
Theano is another open source library learning algorithm (Mila) of the Montreal Institute. In September 2017, yoshua bengio, an industry-leading AI researcher, announced the termination of Mila's development of theano because these tools have become more popular. Yoshua bengio said in a Mila announcement: "the software ecosystem supporting in-depth learning research has developed rapidly and has now reached a healthy state: open source software has become the norm; Various frameworks are provided to meet the needs from exploring new ideas to deploying them into production; Strong industry players are stimulating competition to support different software stacks ".
Today, there are many open source tools for developers to choose from, including keras, Microsoft cognitive toolkit and Apache mxnet.
2. The need for real-time decision-making of edge artificial intelligence pushes AI to the edge. Run AI algorithms on edge devices such as smart phones, cars or wearable devices without communicating with the central cloud or server, so that the device can process the information provided locally and respond faster to the situation.
NVIDIA, Qualcomm and apple, as well as many emerging companies, are committed to building chips specifically for AI workloads at the "edge". From consumer electronics to telecommunications to medical imaging, edge AI has an impact on every major industry.
For example, an autopilot must respond to events in real time and play a role in areas without Internet connection. Decisions are extremely time sensitive, and large delays can be fatal.
Large technology companies made a great leap in edge AI between 2017 and 2018.
Apple released a11 chip with "neural engine" in 2017, which is applicable to iPhone 8, iphone8 plus and X. it claims that it can perform machine learning tasks of up to 600 billion operations per second. It supports new features of iPhone, such as face ID, which can run face recognition on the device itself to unlock the phone.
Qualcomm launched a $100 million AI fund in the fourth quarter of 2018 to invest in start-ups, "sharing the vision of device AI and making AI more powerful and extensive". Its actions are closely related to the 5g vision.
As the leading processor in many data centers, Intel had to play a role and make large-scale acquisitions. Intel released the vision of device AI: the processing chip myriad x (originally developed by movidius and acquired by Intel in 2016).
Intel launched Intel ncs2 (neural computing stick 2) in the fourth quarter of 18, which is supported by myriad x vision processing chip and can run computer vision applications on edge devices such as smart home devices and industrial robots.
CB insights earnings record analysis tool shows the following:
Edge AI is on the rise in part of 2018
Microsoft said that in the third quarter of 18 years alone, it launched 100 new azure functions, "focusing on existing workloads such as security and new workloads such as IOT and edge AI." NVIDIA recently released the Jetson AgX Xavier computing chip for edge computing applications across robotics and industrial iots. Although edge AI reduces latency, it also has limitations. Unlike the cloud, the edge has storage and processing constraints, and there will be more hybrid models that allow intelligent edge devices to communicate with the central server.
3. Face recognition has become the mainstream from unlocking the mobile phone to boarding and flying. In face recognition, China's driving force for security and AI's ambition have become the focus of media attention.
As the government adds a layer of artificial intelligence to monitoring, start-ups play a key role in providing basic technology for the government. Quickly search the transactions of Chinese face recognition start-ups on cbinsights platform, reflecting the demand for this technology.
Unicorns such as sensetime, face + + and recently cloudwalk have emerged from China（ This is our detailed report on China.) However, according to CB insights patent analysis tool, even in the United States, interest in the technology is surging.
Apple popularized this technology for daily consumers by introducing face recognition based login technology in IOS 10. Amazon is selling its technology to law enforcement agencies; Academic institutions such as Carnegie Mellon University are also studying technologies that can help enhance video surveillance, and have obtained the patent of "making facial features clear" - a method that can help law enforcement agencies identify covered suspects by reconstructing the whole face when only the eye area of the face is captured, Then, face recognition can be used to compare the "translucent face" with the actual face image to find the face with strong correlation. But the technology is not without faults. It is reported that Amazon has received much news coverage for mistaking some members of Congress as criminals. When the "smiling face unlocking function" is temporarily disabled, smart cameras outside Seattle schools are easily deceived by reporters of the Wall Street Journal, who use photos of the principal to enter the place“ Smiley unlock and other such activity detection methods provide an additional layer of authentication. For example, Amazon has obtained a patent that can explore more security layers, including requiring users to perform certain operations, such as "smile, blink or tilt their head". These operations can then be combined with "infrared image information, thermal imaging data or other such information" for more reliable authentication. Early commercial applications are developing rapidly in the fields of security, retail and consumer electronics. Facial recognition is rapidly becoming a main form of biometric authentication. 4. Medical imaging and diagnosis FDA is approving AI as a medical device. In April 2018, FDA approved the AI software, which can screen diabetic retinopathy for patients without the need for experts to make second observations. It was awarded the title of "breakthrough equipment" to speed up the process of bringing products to market. IDx-DR software can correctly identify patients with "mild diabetic retinopathy" in 87.4% of the time, and identify patients without diabetes in 89.5% of the time. IDX is one of many AI software products approved by FDA for clinical commercial applications in recent months.
The FDA approved vizlvo, a product of start-up viz.ai, to analyze CT scans and inform healthcare providers of the possibility of stroke. After obtaining FDA approval, viz.ai obtained a round a financing of US $21 million from Google ventures and Kleiner Perkins Caufield & Byers. FDA also approved the oncology AI suite of arteries, a start-up company supported by GE ventures, which initially focused on the discovery of lung and liver lesions. Since 2014, rapid regulatory approval has opened up new business channels for more than 80 AI imaging and diagnostic companies that have raised equity financing, with a total of 149 transactions.
On the consumer side, the popularity of smart phones and the progress of image recognition are turning the phone into a powerful home diagnostic tool. Dip.io, the first product of startup health.io, uses traditional urinalysis test paper to monitor a series of urinary tract infections. Users use their smartphone to take photos of the test paper, and then the computer vision algorithm will calibrate the results to consider different lighting conditions and camera quality. The test can detect infection and pregnancy related complications. Dip.io, which has been approved by FDA, has been listed in Europe and Israel.
In addition, many ml (machine learning) as a service platforms are being integrated with FDA approved home monitoring devices to alert doctors in case of abnormalities.
5. Predictive maintenance
From manufacturers to equipment insurers, AI iiot can save existing enterprises millions of dollars in accidental failures. A lot of data is generated by field and factory equipment,
Our other product:
