PRCV 2020 grand banquet, Pengsi technology shows the application results of transfer learning

category:Internet
 PRCV 2020 grand banquet, Pengsi technology shows the application results of transfer learning


After showing the layout and research results of Pengsi technology in visual aiot last year, at this prcv2020 academic forum, Robert Lorenz, Ph.D. of Humboldt University and chief algorithm researcher of Pengsi technology, delivered the keynote speech of Pengsi Technology: making efforts to transfer learning, striving to build general artificial intelligence, and shared with the guests and scholars attending the meeting that Pengsi technology has made great contribution to general artificial intelligence Intelligent thinking, and the application of visual intelligence algorithm based on transfer learning.

Robert said that general intelligence is the inevitable trend of the development of the next generation of AI, representing the future of the intelligent revolution; transfer learning is the only way to move towards general artificial intelligence. He focused on the important technical achievements of Pengsi in the field of artificial intelligence research and development - Pengsi transfer learning platform. The platform focuses on training the pre training model with good performance on the source task, and then optimizes the model through unsupervised / self supervised learning, small sample learning, Gan, multi task learning and cross domain learning in transfer learning, so as to realize the automatic improvement of AI performance.

As an important branch of pattern recognition and computer vision technology, transfer learning has a wide range of innovative research and application in Pengsis business. For example, in the aspect of pedestrian recognition Reid, Pengsi innovatively combined the confrontation generation network with self-monitoring learning algorithm to train the model. Through migration learning, it carried out the research and development of high-accuracy cross scene (database) pedestrian recognition algorithm, and refreshed the world records of cross domain pedestrian re identification (Reid) market1501, dukemtmmc-reid and msmt17 data sets.

In the cvpr2020cd-fsl challenge held in June, Pengsi technology proposed the metatransfer learning / metafine tuning algorithm model and graph convolution GNN / graphneural network, which were combined into the basic framework of the algorithm, so that it can achieve good performance in four different domains of crop diseases, satellite images and skin lesions, as well as chest X-ray Excellent results.

Since this year, Pengsi has continuously increased R & D investment in transfer learning, unsupervised / self supervised learning, small sample learning and multimodal learning, and has played a good role in industrial application, forming productivity transformation, providing strong technical support for the construction of Pengsi aiot ecosystem. As an international cutting-edge aiot ecological platform company, Pengsi technology will adhere to the international forefront of AI algorithm research, actively promote the implementation of AI technology in practical application, pay attention to emerging new scenes and new demands, let AI serve all aspects of social life, and drive the continuous evolution of cutting-edge AI technology to inclusive intelligent services. Source: editor in charge of mass news: Chen Tiqiang_ NB6485

Since this year, Pengsi has continuously increased R & D investment in transfer learning, unsupervised / self supervised learning, small sample learning and multimodal learning, and has played a good role in industrial application, forming productivity transformation, providing strong technical support for the construction of Pengsi aiot ecosystem. As an international cutting-edge aiot ecological platform company, Pengsi technology will adhere to the international forefront of AI algorithm research, actively promote the implementation of AI technology in practical application, pay attention to emerging new scenes and new demands, let AI serve all aspects of social life, and drive the continuous evolution of cutting-edge AI technology to inclusive intelligent services.