Source: ATYUN AI platform
May 18 to 20, computer vision Turing's top event --ACM China conference in Shanghai come to an end. "Leading artificial intelligence to create endless possibilities" as its theme, this conference brings together the world's most powerful brain from academia and industry, including three Turing Award winner - Google vice president and chief Internet experts Vinton Cerf, Harvard University Leslie Professor Valiant, Carnegie Mellon University, Raj Reddy, and machine learning dean Michael I. Jordan, sun media Group chairman Yang Lan, vice president of Baidu Wang Haifeng, iFLYTEK CEO Hu Yu and so on. Long code technology and the honor of the world's leading experts, scholars and entrepreneurs on the same stage, to discuss the application of cutting-edge technology in today's society.
Association for Computing Machinery (ACM) was established in 1966, "Turing Award" to reward individuals who have made important contributions to the international computer industry, and to commemorate the "father of artificial intelligence" Alan Mathison Turing (Alan Mathison Turing) great contributions to computer science. The award for the computer industry's most prestigious and highest honor, known as the "Nobel Prize computer industry," he said. Following last year's ACM Turing Award of the fiftieth anniversary of the General Assembly was successfully held in China this year, China Turing Assembly once again held in Shanghai, invited many guests luxury academia and industry, innovation and development around the areas of computer technology industry a plurality of topics such as applications started to explore.
During the meeting, Dr. Wei-Lin Huang Lung Technology Chief Scientist code published to "Computer Vision in RetailAI: recognition from the object to identify the goods" in the title of the keynote speeches, the guests and scholars, entrepreneurs and students to explore technological breakthrough computer vision and commercial applications.
Dr. Wei-Lin Huang was a postdoctoral fellow at Oxford University Visual Geometry Group (VGG) laboratory, under the tutelage of Andrew Zisserman and Alison Noble, mainly in the areas of research scene text recognition, scene classification and medical video analysis during the post-doctoral.
Dr. Wei-Lin Huang issued to: Keynote Speech "Computer Vision in RetailAI recognition from the object to identify the goods" in the title of
Speech, chief scientist Dr. Wei-Lin Huang briefly introduced several important milestones in the development of computer vision, comes to product identification, as well as technical problems currently encountered in real business scenarios from the image classification, and shared how long science and technology through the original code weak supervision with the algorithm to technological breakthroughs.
First, a brief introduction Dr. Huang ImageNet image recognition competition for computer vision development has an important contribution. In 2010 and 2011, when everyone uses the traditional method of using same error rate above 25%. Until 2012, the convolution neural network was first applied to image recognition ImageNet game, making the error rate decreased from 25.8% to 16.4%, a huge performance boost. This breakthrough led to changes in computer vision technology, from the depth of learning technology is widely used throughout the field of computer vision. 2017, the error rate down to 2.2 percent, exceeding the 5.1% error rate performance of human recognition, proved in 1000 Image classification this task, the machine can do better than humans.
However, the product identifiers long science and technology focus in this area, identified as many as hundreds of thousands of species, because even the same product, different brands or models of different prices are not the same, we need to do SKU (stock quantity units) classification level. It can be seen, the commodity recognition technology in a real-life product image recognition more difficult, much more difficult to identify than 1000 kinds ImageNet game objects.
It is noteworthy that, ImageNet contest for visual image recognition technology has played a significant contribution to the development of the computer, with large-scale data sets that manual tagging is critical. However, due to the high cost of manually labeled data, more and more researchers began to focus on using low-cost data (such as data without manual annotation) to train the way an image recognition system.
Long code technology by weak supervised learning algorithm independent research and development to solve the problem of manual annotation. Dr. Wong layman's language shared weak supervised learning courses and training strategies how to break through technical difficulties depth study required a lot of manual annotation data to support. Supervised learning algorithm by weak original, Dr. Huang led the team not only deal in commercial real scene of massive noise data, reducing the enormous costs of manual annotation, and therefore also won the large-scale visual understanding WebVision World Challenge on CVPR 2017 first place award. The results of this race also shows that even in more in line with actual usage scenarios, without manual cleaning and annotation data, the algorithm code Long Technology team achieved 94.78 percent accuracy rate performance (equivalent to an error rate of close to about 5% ), also has a human can reach par performance.
Dr. Wong also shared the practical application of the current long code identification technology in new technology goods retail electricity supplier, garment and textile, household furniture and other industries, to get the guests alike. At the awards dinner, Dr. Wei-Lin Huang received excellent Keynote speakers certificate by the Turing Award winner Dr. Vinton Cerf personally issued.
In the booth venue, yard-long science and technology have the opportunity to communicate with many outstanding scholars and students, and to demonstrate the application of its core products ProductAI in new retail electricity supplier, garment and textile, furniture, household and other industries. Among them, "Smart Container purely visual product identification solutions" was the presence of experts and scholars, teachers and students attention. The program can help get rid of the cost of traditional retail container shackles of gravity sensing module and RFID consumption and efficient solving pain points within the new retail scene business side of operations and consumer shopping. Landing of this technology is based on the ability to code product identification technology core products ProductAI Long artificial intelligence product identification platform high precision, the algorithm behind the code long science and technology training team in huge amounts of data, accurate tireless efforts on algorithm model design.
The event, yard-long science and technology with the world's elite gathered in this Chinese Turing Assembly, jointly outlined a blueprint for the future of artificial intelligence. Future, yard-long science and technology will continue to focus on computer vision technology innovation and breakthroughs, and actively, in close cooperation with academia industry, will be at the forefront of academic trends, the latest research results presented to more committed to innovation and scientific workers.
This switched ATYUN artificial intelligence media platforms, the original link: Turing the General Assembly come to an end, Long code technology and the world's strongest brain on the same stage
more recommendations
Google launched artificial intelligence end to end platform for building custom models
BenevolentAI: the use of artificial intelligence to optimize small-molecule drug design
In this study, academic publishers publish books first generation of AI
Separable convolution basic introduction
