Algorithm Powers Innovation

Reinforcement learning for combinatorial optimization problem is the hot topic in academic research. Resource scheduling on the cloud is a typical combinatorial optimization problem in the cloud scenario. We will work closely with professors in ECNU on this topic to explore the application future of reinforcement learning in industry.

Based on the DNN-HMM framework acoustic modeling technology, the speech algorithm team successfully landed on a new model of 16K Real-time speech recognition continuous mode at the CN North Beijing4 region. It can recognize more accurately recognizing Mandarin while some simple dialects.

On May 28, 2020, Prof. Dai Wenrui, from Shanghai Jiaotong University, was invited by CLOUD BU Algorithm Innovation Lab to give an academic report on "Learning-based Image and Video Compression Methods". This report is broadcasted online through WeLink software, and has attracted more than 110 colleagues in Huawei to participate in.This report mainly discusses the design of end-to-end video and image coding schemes inspired by traditional image/video framework and methods. Prof. Dai introduced traditional image and video compression methods based on dictionary learning and current end-to-end deep learning methods. He also compared the advantages and disadvantages of various learning-based image/video compression methods from various perspectives, as well as future video encoding and decoding technologies. In the interactive session, Prof. Dai gave wonderful answers to the questions that everyone cared about.Many Thanks to Prof. Dai for the wonderful report!

Dr. Zhou, the expert of the Resource Scheduling Algorithm Team, was invited to make a report of "The Resource Scheduling and Intelligent Algorithms Behind Second-Scale Expansion of Alkaid " on Alkaid TechTalk's live broadcast.

Some key members of the Resource Scheduling Algorithm Team had some technical discussions on the topic "Resource Scheduling Algorithm on Cloud Computing" with some professors and researchers from Huazhong University of Science and Technology, Northwest Polytechnic University, Xi'an Jiaotong University, East Nanjing University, and East China Normal University.

The Intelligent Caching and Risk Management Algorithm Team incubated a new service of intelligent verification code checking based on AI and security, built a significant competitiveness on human-machine confrontation and achieved precise defense from multi-platform and multi-type machine attacking behaviors. In advanced anthropomorphic attacking scenarios, the actual network detection rate is notably higher than other companies.

Content review algorithm was successfully applied on politician and terror detection service, the F1-Score of which surpassed other companies. After multi-round competition with TOP competitors, we ranked the first in several tests, including the terror identification service, politics-related service and celebrity identification service, successfully took the largest commercial cloud customer in the XX live POC. The monthly call rate reached over several billion times and the total call number was over tens of billion.

Voice classification service was put into commercial use at the automatic learning platform of ModelArts. Its precision and recall rate are top on the public dataset. The service was applied to the Tech4All Rainforest Project, to help zoologists better protecting rare animals by detecting illegal cutting based on the technology of detecting the sound of electric saw.

The rate of voice wake-up (true positive rate) was greater than 95%, and the false wake-up rate (false positive rate) was less then 1 time / day. The performance of the service reached the industry leading level and was launched on The HUAWEI CLOUD ROC assistant.

The Intelligent Caching and Risk Management Algorithm Team incubated a real-time intelligent defense platform and intelligent anti-fraud new service based on AI + security, and built the strengths of innovation, analysis ability and others through the edge-cloud collaboration and multi-mode memory fusion analysis architecture. By providing HUAWEI CLOUD with a daily fraud detection of xxx million, the service successfully intercepted many black-product attackers. In the same year, the product service was released to the world at the HC2018 conference.

Cry recognition was successfully applied to the Puffin AI panoramic camera and launched on the Hilens skill market.

We successfully incubated big data real-time analysis engine, developed 10+ algorithms about graph mining and spatiotemporal analysis, which were launched on HUAWEI CLOUD Service CloudTable.

The Intelligent Caching and Risk Management Algorithm Team incubated an intelligent ultra-high-dimensional AI modelling decision-making products based on Fusion Insight RTD, and reserved the super-large (millions) concurrent real-time decision-making technologies. Real-time intelligent decision-making service has been widely used in various business lines of Bank Z, including credit card risk management, retail risk management, credit risk management, etc., supporting Bank Z's 800,000 computations of one million dimensional risk and AI model in one minute, which made Bank Z winning the financial innovation award (Shanghai Financial Innovation Award http://www.shanghai.gov.cn/nw2/nw2314/nw2319/nw10800/nw39220/nw44391/u26aw62816.html) and taking the leading position in this field. Our sales volume in the field of financial risk management reached X billion and got the Thank You Letter form the client.

The Fusion Insight RTD product incubated by the intelligent cache and risk management Algorithm team was deployed on the Huawei Consumer Cloud, handling xxxxx million risk requests daily, which greatly improved the risk management capabilities of consumer cloud accounts.

The Intelligent Caching and Risk Management Algorithm Team incubated Fusion Insight RTD products and financial risk management solutions, which were released at the HC conference. This is the first time that our products are integrated into the bank's core IT system, replacing the industry benchmark ACI PRM and mainframe.

In the XX public security project, the relationship analysis algorithm was optimized to solve the performance problem of the relationship algorithm when the expansion rate was greater than xx times, making the number that the relationship discovery algorithm can support increased by x times .