On July 27, TeaTalk · The second issue of the Online Application Practice Series - "AI Technology, Simplifying the Complex World" was successfully held. In this live broadcast, we followed Teacher Yang and let AI technology meet the mobile cloud. By understanding how AI technology monetizes technology and how to apply it quickly on the cloud, let's unleash our imagination and spark endless possibilities.
The following is the transcript of yang Xi's speech, a senior algorithm engineer at China Mobile Cloud Competence Center.
The development of artificial intelligence technology has now been relatively mature, and now artificial intelligence technology has gradually matured from an emerging cutting-edge technology to an inclusive technology. At the same time, how to realize the realization of technology by today's artificial intelligence technology and how to achieve rapid application of artificial intelligence technology on the cloud are the hot spots.
This sharing mainly includes several aspects:
- Explain the origins of artificial intelligence technology and mobile cloud
- Introduce typical cases of common API-like artificial intelligence technology applications
- Introduce typical cases of common platform artificial intelligence technology applications and related technical characteristics
- Summarize and prospect the application of artificial intelligence technology on the cloud
China Mobile Cloud Competence Center began research on artificial intelligence technology in 2014 and has accumulated for 8 years today. From the perspective of the development process, the whole is divided into two stages:
Stage 1: From 14 to 19 years, we will mainly build artificial intelligence-related projects for the provinces in the mobile system, including spam SMS projects, legal contract governance projects, and so on.
Stage two: From the beginning of the cloud reform in 19 years to the present, the accumulated artificial intelligence capabilities will be upgraded to the mobile cloud, and at the same time, easy-to-use and easy-to-use artificial intelligence products will be created.
About mobile cloud artificial intelligence technology
The overall layout of mobile cloud artificial intelligence takes mobile cloud infrastructure as the carrier to create an intelligent engine of cloud + AI. Mobile Cloud independently develops AI platforms, provides AI learning platforms, AI capability services, and AI industry applications, and provides rich AI services from the three levels of basic platform, core capabilities, and application products for various vertical industries and industrial chains. The current mobile cloud-cloud integration product system provides a total of 23 products, covering all fields of text, image and speech processing.
The mobile cloud artificial intelligence product system provides a full range of AI products such as hardware, learning platform, capability services and application services from the bottom up.
- The basic resources of the mobile cloud include containers, microservices, CPUs, GPUs, etc., providing a basic computing storage environment for all business and intelligent applications.
- The AI learning platform includes an intelligent interaction platform, a deep learning platform, a visual modeling platform, and a knowledge graph platform, which support customers to quickly complete model development and service deployment, and accelerate the AI application launch process.
- AI Capability Service provides a total of more than 50 open capability APIs covering the three major technical fields of natural language processing, machine vision, and intelligent speech, with the advantages of rich capabilities, leading technology, efficient and stable, and easy to use services.
- Mobile cloud AI industry applications combine the actual business conditions of various industry applications, such as face cores and smart communities, to provide industry-wide solutions.
About how to use mobile cloud AI products
There are many types of mobile cloud artificial intelligence services, from the degree of difficulty of use, the way of interaction, can be roughly divided into three types of AI application services, which can be integrated into different products or corresponding application environments in different ways.
- API products: Usually only a small number of API interfaces are provided for interactive use, and the console interface only has simple usage statistics, and it is impossible to personalize the relevant capabilities, and the related capabilities are basically generalized capabilities.
- AI Light Platform Products: Unlike API products, the console of this class of capabilities can be easily configured. Different users can use different configuration properties, which can be more personalized than API products
- AI-heavy platform products: In addition to configuration management, the use of capabilities is very dependent on the visual interface provided by the platform. It is also more difficult to use than the first two. For example, deep learning platforms, interactive platforms, and so on.
At the moment when epidemic prevention and control is normalized, health codes have become a necessary credential for everyone to pass. Combined with OCR, the use of named entity identification can quickly and efficiently automatically identify and verify the health code, so as to ensure the rapid verification of everyone's access to public places, and greatly reduce the workload of epidemic prevention personnel. Mobile cloud from 2021 to support the identification of big data itinerary card, epidemic prevention health code, after more than a year of development, the formation of a relatively perfect epidemic prevention health code identification program.
The use of the overall capability is also very simple, you can select the corresponding product of artificial intelligence in nine days in the mobile cloud, order the corresponding artificial intelligence product, and then enter the mobile cloud console to generate AK and SK for interface authentication. Through the interface address of the Help Center, the interface is accessed through the Internet. Verify the interface call through the feedback results of the interface.
Usually API-like AI products only provide interfaces to Restful APIs, and basic development and integration capabilities are required to use such capabilities. In general, such capabilities are used as underlying atomic capabilities, embedded in application products. For example, the identification of the epidemic prevention health code can be docked in the cloud space using the interface to complete the function of health punching. And can actively detect and manage the health status of employees.
In addition to the epidemic prevention health code, the mobile cloud also provides a responsive AI capability interface in various scenarios, such as in the OCR capability scenario, in terms of ticket recognition, the mobile cloud provides rapid identification of train tickets, fixed invoices and other bills, and in terms of cards, the mobile cloud provides the identification of common cards including ID card recognition and bank card recognition, which can achieve rapid access to the interface.
In terms of light platform applications, the main introduction is the content moderation product, the content review product is a full-stack intelligent content recognition product, support video, audio, documents, pictures and other content review, accurately identify multi-scenario illegal content, efficiently defend against content risks, improve review efficiency. Content moderation combines some of China Mobile's experience in spam SMS and web phishing website identification, abstracts the AI atomization capabilities in it for platformization, so that customers can achieve personalized identification of a variety of illegal content through simple configuration.
Rich media content review products can intelligently detect diverse scenarios of video, audio, documents, pictures and other objects, effectively reducing the cost of manual review, improving the efficiency of review, and greatly reducing the risk of content violations.
- Intelligent audio moderation: Review audio files in a variety of formats, and support the identification of pornography, violent terrorism, political sensitivity, advertising, spam, etc.
- Intelligent video review: The image and audio of video files are reviewed in two dimensions, and support the identification of pornography and vulgarity, bad scenes, bloody and violent terrorism, gun knives, garbage advertisements, sensitive characters, etc.
- Intelligent image moderation: Identify pornographic content, flag logos, gun knives, bloody riots, terrorist organizations, spam ads, bad scenes and other content in pictures, and support user-defined black-and-white gallery.
- Intelligent document review: Support word/pdf/ppt/zip and other documents in the image, text content for review, support pornography, violent terrorism, political sensitivity, advertising, spam information and so on identification.
At the text content review level, it supports the identification of multiple types of spam, sensitive words and variants such as violent terrorism, pornography, vulgarity, irrigation, and advertising in the text, and supports the configuration of custom sensitive words to help customers improve the efficiency of review and avoid business risks.
- Article content review: It can identify pornographic, reactionary, vulgar and other content for articles submitted by users/authors, so as to avoid the platform from distributing articles containing illegal and illegal content, and reduce the risk of content production platforms being punished by supervision.
- User review filtering: Can detect the website user comment information, once found that the user publishes comment information involving pornography, vulgarity, abuse and other content, it can be filtered in real time to ensure a good user experience of the product.
- Live content review: It can identify and filter content involving pornography, vulgarity, abuse and other content in the bullet screens, comments and other remarks sent by users in live broadcast scenes, and purify the Internet content environment.
Intelligent image moderation can identify pornographic content, flag logos, gun knives, bloody riots, terrorist organizations, spam ads, undesirable scenes and other content in pictures, and support user-defined black-and-white gallery. Common usage scenarios include avatar review, image review with heavy comments and messages, and image review for scenarios such as cloud albums.
Intelligent document parsing product is a document structured parsing product that integrates labeling, model training and inference. Users only need online annotation to quickly customize the exclusive model, which is a zero threshold, high efficiency and high value product. The extraction and review of key elements of alternative manual processing texts can help enterprises solve the problems of long document review, labor costs and high risks, and realize the internal processing process intelligently.
Intelligent document parsing has five major functions: multi-person writing, text parsing, online annotation, model customization, and online verification, which can quickly meet the personalized analysis needs of users for the analysis of document products.
Taking a co-inspection application as an example, due to the need to review the user's qualifications, it is necessary to provide the service. Therefore, every time the user presents a business need. Often need users to upload relevant certification documents and the system reported by users in the system for comparison, before using the intelligent document analysis products, basically can only use manual comparison for the relevant documents to compare, the efficiency is slower not to say, for non-office hours there is no way to verify. After using the intelligent document parsing product, 7x24 hours of automated document audit can be realized, which greatly improves the efficiency of document audit.
From the perspective of using document parsing products, we only need to provide document samples of related files (pdf, word and jpg picture formats are currently supported), and then through the interface operation, we can realize the definition of relevant field properties, as well as the rapid annotation of data, without understanding the training process of the model, we can realize the automatic training and release of the model. Quickly extracting fields such as mobile phone number and residential address from 1,000 pieces of data can be completed in a matter of minutes, and the overall analysis efficiency has been greatly improved.
Summary and outlook
1. Providing common AI services through the cloud has become a common consensus and trend. The integration of mobile cloud and cloud intelligence has empowered hundreds of industries and improved industrial efficiency.
2, the unification of AI models, the big model is the future trend of AI applications, ai big models can provide more convenient application effects on the basis of less prompts/data, and large models can provide services through the cloud to better save costs, and the cloud can become the standard infrastructure for future applications of large models of AI.
Scan the code to view the event review and expert PPT
This issue of TeaTalk · Online We followed Yang to understand how AI technology liberates productivity and simplifies a complex world, and we hope we will accompany you through an imaginative night. As an important part of the developer community series of activities, The Future TeaTalk· The Online Live Streaming column will focus more on subdividing the technical field, further expand the breadth of knowledge and technical depth, and accelerate the embrace of developers.
Stay tuned for us, we'll see you in the next issue!
We are invited to join the Mobile Cloud Developer Community
Build a mobile cloud ecosystem together!
Scan the code to enter