Traditional industry application AI faces three major challenges of high quality data is the application premise

Traditional industry application AI faces three major challenges of high quality data is the application premise

Original title: Traditional industry application AI faces three major challenges of high quality data is the application premise recently, Wen Shanta, a well-known artificial intelligence scholar, published an article, explaining his slow understanding of artificial intelligence in traditional industries.

Whether it is a personalized recommendation when brushing short video, or time-consuming time-consuming time, or the face recognition of mobile payment, Ai technology represented by algorithm is "well-handed" in the consumer Internet industry. However, it is difficult to quickly think of a typical case of very mature application artificial intelligence. Why is AI technology to apply the traditional industry to the consumer Internet and other industries? Consumer Internet Industry Application AI More Advantages "AI Technology Application Mainly depends on data, intelligence and algorithm." Zhu Pengfei, associate professor, Associate Professor of Tianjin University, First, the data should be a certain amount of volume, which is the basis of application, in addition The integration must support large-scale model training, and the post-algorithm needs to reach a certain accuracy, and the end side must have certain reasoning ability.

The reason is currently only consumer application AI technology in large-scale applications, mainly in these three aspects of consumer Internet companies.

The short video of the previous year did not now be hot, such as the initial Taobao, and there is no strong user sticky.

As the push is getting more accurate, the user’s experience has also been greatly improved, and finally a well-vented user growth.

"Precision Push mainly depends on the improvement of algorithm accuracy, and the improvement of algorithm accuracy is inseparable from massive data." Zhu Pengfei explained that in this single scenario, the algorithm model needs to evolve, lifetime learning.

Since it is not a closed data environment, there is always a new data to join, the algorithm model needs to be continuously adjusted, iterative upgrades, making its accuracy getting higher and higher to form a benign loop. At the same time, although the current consumer Internet industry has risen to a certain height in algorithm accuracy, the threshold for consumer Internet industry for AI algorithm accuracy is relatively low compared to some traditional industries. For example, short video, Taobao Preferences recommendation, Baidu hot search keywords, only need to meet the purpose of the user, as long as there is certain accuracy, the user can accept.

"Zhu Pengfei said, in contrast, in many traditional industries, the requirements for technical precision are much higher.

For example, visually based AI technology applied to face recognition, in high-speed rail stations, airport verification identities, 1: 1 comparison accuracy is up to% or even higher to apply.

In terms of interactive, the current cloud integration has supported large-scale model training and reasoning, such as short video, Taobao recommendation, etc.

However, in a large number of traditional industry application scenarios, the end side integral force on the intelligent terminal can not meet the real-time and accuracy requirements of reasoning. "Compared to social networks and e-commerce systems, the closed ecosystem of traditional industry application scenarios makes cloud integrity unable to apply effective applications." Zhu Pengfei said, with intelligent unmanned system inspection as an example, power inspection, pipeline inspection , Transportation inspection, river inspection, and photovoltaic inspection requirement to carry a real-time inspection requirement to meet the real-time inspection requirements on the drone and robot. Due to the high complexity of video analysis, the end is often unable to achieve accurate and efficient real-time real-time. The reasoning, lightweight networks lose their recognition accuracy while meeting real-time.

Since the accuracy of the algorithm is not available, the application of AI technology cannot be achieved in many scenes.

Traditional industry application AI faces three major challenges, Wuanta believes that in Ai applications, other industries outside the consumer Internet industry face three challenges: the data set is small; the customization cost is high; the process of verifying the idea to deploy production very long. In this regard, Zhu Pengfei also deeply feels, and he analyzes the traditional manufacturing industry as an example.

"Traditional manufacturing companies are a very prominent issue in the process of manufacturing the transformation into the wisdom.

Zhu Pengfei introduced that there is a certain difficulty in the acquisition of data. The data of traditional manufacturing companies is closed because many traditional enterprises are not new informatics, and there is no sensor to collect real-time data, and there is no data center, so data zeroat, missing Serious, it is difficult to obtain the kind of massive, high-quality data like consumer Internet enterprises. Secondly, there are many commercial value of various factories in the industry, so the factories are strict, which caused data without flow, there is no way to share, and therefore The data is island effect has affected the optimization of the AI ??algorithm model. "When developing an AI algorithm model, because the data is confidential, the data that is often obtained is ‘desensitive’, which has also seriously affected. Our judgment.

In the traditional industry, there is a lack of technicians with the AI ??algorithm model development capabilities, so both parties have a very high barrier in cooperative research and development.

Zhu Pengfei said. In addition, data sources in traditional industries are not like consumer Internet fields. The complex business scenes caused data to be "dirty", and "cleaning" must be "cleaning", and the AI ??algorithm model can High efficiency learning to increase accuracy.

"This is like teaching children’s knowledge, just telling knowledge points, children can learn quickly, if they mix a lot of useful information in the knowledge point, children are not distinguished, and learning efficiency is definitely lowered." Zhu Pengfei introduced, and give data label The "knowledge point" work is huge and cumbersome, requiring companies to do it, spend a lot of time energy. "Traditional manufacturing needs to obtain high quality data, it must be transformed into information and intelligent production equipment." Zhu Pengfei said that this transformation requires enterprises to invest a lot of time and energy, and will increase production costs, which is also Become a barrier for AI application in traditional manufacturing. High quality data is the application premise in the past 10 years, most AI’s R & D and application is "software-centric" driven.

Under the support of massive data, the software and algorithms are continuously optimized to achieve higher algorithm accuracy. In the case where the traditional industry cannot improve the quality and quantity, Wu Yida believes that the traditional industry should adopt the "data-centered" mode, put the focus on data that is better, higher than the match. "Under this idea, traditional industries have emerged in some good application cases.

For example, the image recognition AI system in the medical field can help the doctor ‘to see the’ CT image film, identify the lesions such as tumors, and assist the doctor to make judgments.

"Zhu Pengfei introduced that because many data were labeled by professional radiologists on the video film, the data is relatively accurate, the AI ??algorithm model has improved in the process of learning. At present, many image recognition systems can reach More than 90%, since it is an auxiliary doctor, and finally, the doctor needs medical decision, but the accuracy of this level has largely reduced the work strength of the doctor.

"Although the traditional industry has some successful cases of application AI technology, we must better combine with AI, but also have to improve data quality.

"Zhu Pengfei suggested that the traditional industry that has already accumulated massive data, under the premise of ensuring data security, the value of the data. The value of the data is excavated, and the demand is associated, there will be a lot of development space. Secondly, for Emerging industries, such as new energy vehicles, etc., when building intelligent factory planning, the factors that get data and intelligence are taken into account.

However, Zhu Pengfei emphasized that while using Ai technology in traditional industries, do not abuse Ai technology, do a good job before application, if production efficiency is not increased, the industry has improved, so blindly uses AI technology, that is, resources Waste. "For example, some application scenarios need AI algorithm to reach more than 99% of the accuracy to use, by evaluation, existing model algorithms can only reach 90% accuracy, then this scene is not necessary forced to force Ma Ai technology.

"In summary, for the application of AI technology, it is necessary to talk, there is high quality data, and there is no good data. It is difficult to apply.

Zhu Pengfei said. (Chen Yu) (Editor: Mouquet Yu, Zhu Hongxia) Share more people to see.