Shulou(Shulou.com)06/03 Report--

In the first part of Autumn Ripples of the Industrial Revolution, we review a series of questions: Why does China need an AI-driven industrial revolution today? And why nurture this revolution? On this basis, we discussed the core position of the deep learning framework and how Baidu can make the flying propeller go to the industrial level and become the operating system of the AI era.

If this series of logic helps us fully describe the propeller as the engine of the AI industrial revolution, then we should ask how the power of the propeller is transmitted to the various parts of the machine to truly make the behemoth of the industrial world work.

In this chapter, let's straighten out the order of "guiding method-technical logic-industrial case," and answer the question of how to fly oars into reality and how thousands of industries can really stimulate industrial change through deep learning.

The reason why we can find the answer is that on October 17, Baidu hosted the 2019 Zhongguancun Forum·Deep Learning Technology and Application Innovation Forum in the AI Era. Among the Zhongguancun Forum, a repertoire reserved by China's scientific and technological community, this forum is the only parallel forum this year focusing on interpreting the industrial application of deep learning.

Under the watchful eye of deep learning developers, industrial ecological participants and all walks of life, the ripples of industrial revolution stimulated by flying oars have opened a new round of shocks facing the industrial world.

The Road to Technology Base: Wang Haifeng's Four Judgments on Intelligent Revolution

What is a flying paddle? Why do we need flying oars today?

In his forum speech, Wang Haifeng, director of Baidu CTO, Deep Learning Technology and Application National Engineering Laboratory, answered this question.

Back to the brilliant history of industrial revolution of human civilization, Wang Haifeng proposed that the first few industrial large-scale production of mankind required very complex application systems, which must have the characteristics of standardization, automation and modularization at the same time.

According to this logic, in today's AI-driven industrial revolution, there must also be such a system for deep learning to be applied to all walks of life. Among Baidu's goals, Flying Propeller is such an AI operating system that can standardize, automate and modularize deep learning technology. It can be applied to industrial large-scale production in all walks of life, helping the intelligent transformation of the industry and becoming the technical base of the AI era.

So from the deep learning framework to the AI technology base and the industrial intelligent revolution engine, how can the flying propeller achieve its own external release? How exactly do the ripples we're talking about grow outward from the inside of the paddle? In Wang Haifeng's view, there are four key words that must be adhered to when flying oars move towards industrial practice:

1. Open source and opening up. By maximizing the technical accumulation of Baidu and the technical characteristics of flying oars, the industry threshold will be lowered.

2. Autonomous and controllable, Flying Propeller is Baidu's AI operating system with completely owned intellectual property rights. Under the new international background, intelligent support for China's industry can play a key role.

3, soft and hard integration, want to move towards industrial practice, flying propeller must be compatible with all kinds of chips, applications, hardware environment. Therefore, we saw that Flying Propeller was actively carrying out various soft and hard optimization, such as exploring NPU optimization together with Huawei. This is the key power point for the large-scale industrial production stage reserve.

3. Industrial ecology, Baidu and Feioar cannot complete all tasks in AI era alone, but must build ecology together with industrial ecology, industry, university and research circles to construct an organic system of talent-technology-application.

These four judgments can be regarded as the symbol of flying oars to the industrial world. With key technology breakthroughs and clear guidelines for action, Feijiang has embarked on a journey to explore the unknown world of industrial intelligence revolution in various industries.

Compressing 108,000 miles into a somersault cloud: industrial practice of flying oars

Let us think about such a question: Why is it that in Journey to the West, the same goes to Xitian Leiyin Temple, Tang Priest and his disciples have gone for more than ten years, and Sun Wukong himself is always going back and forth? The obvious problem is that Sun Wukong has a somersault cloud, and the same goal can be accomplished in an extremely simple way.

In today's world of industrial intelligence, applying deep learning to core productivity categories is a common goal for countless industries. So the difference is, do you fly over the clouds or walk over them?

Perhaps the value of flying oars can be described in one sentence: technology leadership to reduce development costs; ecological empowerment to lower entry barriers. With flying oars, enterprises can greatly shorten the development process of deep learning models, reduce the difficulty of development, and use Baidu's condensed technical capabilities to promote their own industrial path. At the same time, the ecology of flying oars can also help the growth of talents in the industry and further help industries lower the entry threshold for deep learning.

In the forum, Wu Tian, Executive Director of Baidu AI Technology Platform System and Deputy Director of National Engineering Laboratory of Deep Learning Technology and Application, accepted the flying paddle and its industrial practice in detail, summarizing how to complete the role of "somersault cloud" for us.

Taking OCR recognition as an example, when using rules and machine learning methods to solve OCR problems, the whole process needs to be divided into a series of processes such as detection, segmentation, recognition, decoding, post-processing, etc., requiring a large number of expert support, and using many heuristic features. After entering the deep learning stage, the OCR system can use big data for training, and the use of deep learning features can get very good recognition results. A step forward, Baidu brain and flying paddle integrated end-to-end OCR recognition system, users can call the deep learning rules under the high-precision OCR ability anytime and anywhere_the complex and long exploration process in front, are omitted under Baidu's efforts.

Such examples show that in the whole process of development, training and deployment of AI sub-technology categories and deep learning applications, the work that Flying Propeller will be able to explore and realize beforehand has been completed to the maximum extent, and platform packaging has been carried out. Enterprises and industry users only need to "leap into Lingshan" on this basis to be able to land in the era highland of intelligent revolution.

In the process of deep learning and industry integration, another problem is the scarcity of talents, especially industry talents and AI talents are often incompatible. To this end, Feijiang provides a large number of developer empowerment and talent development programs. For example, it has built an active developer community, launched AI high-end talent training programs such as Huangpu College, and AI fast lane for enterprise deep learning combat camp, etc.

In a series of ecological enabling plans, Baidu also actively publicly announced its own technical system and gained experience to help industry users gain first-hand experience. For example, Wu Tian, on behalf of Baidu Research Institute, released the White Paper on Baidu Brain AI Technology Achievements, which carried out a panoramic analysis of Baidu brain.

Through internal compression technology, external prosperity ecology, flying propeller launched intelligent navigation facing the industrial universe. Today, we can already see the ripeness of the journey everywhere.

From Application to Innovation: Deep Learning Transformation for Thousands of Businesses

How to make deep learning "wake up" in all walks of life, perhaps in addition to technical logic, must be accompanied by specific cases to understand.

In fact, flying paddle has been widely used in Baidu inside and outside, all walks of life countless scenes. Perhaps the most intense cross-industry demonstration of the universal AI base capabilities of the Flying Propeller.

For example, on Baidu map, I think everyone should have found that the experience of Baidu map today has far exceeded that of competitors, and its key is the word "intelligence." For example, on the intelligent prediction scene of travel time of Baidu map, based on the flying paddle platform, Baidu map improves the convenience and training efficiency of AI model development and debugging, and completes the training of ten billion data at the sky level on the basis of Baidu map data. In addition, on the support of the flying paddle, it has realized the ability to call tens of billions of times to the sky level, increasing the accuracy rate of user travel time estimation from 81% to 86%, bringing extremely significant efficiency improvement to users.

In the case of intelligent inspection of substations cooperated by Feijiang and Guangdong Power Grid, AI can read accurate numbers on substations with the support of Feijiang platform, and the accuracy rate can reach 99.01%. This liberates the substation digital scribes in remote areas, and the wisdom of AI completes the deep transformation of industrial efficiency and makes human work better. At the same time, the relevant scheme can achieve the completion of the work within 10 days when the new table is adapted, and the installation efficiency is very significant.

And here's another one you might not expect. Golf courses occupy a lot of land and have been a hot topic in previous years. In response to this controversy, it is important to monitor golf course land. The AI land use detection of golf courses is realized by applying the Faster R-CNN model provided by Feijiang. This project builds a national standard golf course remote sensing data set, which improves recognition efficiency by about 90 times.

In the field more related to the national economy and the people's livelihood, flying oar assisted forestry pest monitoring landing is a representative case. The red bark beetle is a stem borer that damages more than 35 species of pine trees. Based on the flying propeller platform, Beijing Forestry University cooperated with Baidu to identify red fat insects through the flying propeller target detection model YOLOv3, realizing hourly data transmission and detection. Compared with manual detection performed once a week, the detection efficiency was greatly improved.

At present, flying propeller has been widely used in industry, agriculture, service industry and other industries. Flying propeller effect is driving the deepening and extensive development of industrial intelligence. From the perspective of industrial users, it is not difficult to find that today we should use this way to quickly realize industrial intelligent practice with the help of flying oar dividends:

1. Move towards the deep integration of deep learning and industry, and build multi-modal, multi-technology integration and soft-hard integrated solutions. Make full use of the technological leadership and industry-level breakthrough of Feijiang.

2. Pay attention to both talents and technology. Make use of the ecological empowerment and talent training of flying oars to actively promote enterprise talents to embrace the AI era and reserve future-oriented development and application capabilities.

3. Relying on ecology and integrating into ecology. Deep learning application is a systematic project that requires division of labor and cooperation in the industrial chain. Actively explore the developer ecology and industrial chain behind the propeller, mobilize the continuous iteration cycle of integrating AI technology and products into its own industry, and reap maximum dividends in the era of industrial intelligence.

So far, we have reflected the overall logic of the intelligent revolution in various industries driven by flying propellers from three perspectives: method, technology and case. At the outer layer of industrial practice, AI ripples that oscillate in circles are affecting the future of the social economy as a whole and the road of human technology.

The light and shadow in the distance, how to merge with the flying paddle to create the kiss of the times, let us stay in the next article to reveal the final secret.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.