More than 100 classic papers were published, and t

  • Detail

On August 28, 2020, the Huawei cloud AI Dean summit was held in Qiandao Lake, Hangzhou. Many academicians from the Chinese Academy of Sciences and the Chinese Academy of engineering, as well as Tsinghua University, Peking University, Zhejiang University, Shanghai Jiaotong University, Fudan University, University of science and technology of China, Harbin Institute of technology, Nanjing University The presidents, professors and experts of AI and computer colleges from more than 40 top universities in China, including Xi'an Jiaotong University, attended the conference. The summit lineup can be called the top configuration of domestic universities

at the summit, Huawei cloud released five AI talent training plans, among which the reproduction of 100 classic papers of Huawei cloud AI 100 academic plan has made significant progress so far: Huawei cloud officially released more than 50 AI classic papers for participants to reproduce, covering computer vision, transfer learning, natural language processing and other fields, attracting dozens of well-known college students to sign up for thesis reproduction, not only covering C9 colleges, There is also no lack of post doctoral participation; A total of more than 100 classic papers and algorithms have been launched into Huawei cloud AI gallery for developers to learn and use. Its variation values are △ R1 △, R2, △ R3, △ R4

more than 100 AI classic thesis algorithms have been launched AI Gallery

in view of the current uneven code quality of AI classic thesis algorithms in the open source community, many codes are not easy to use, difficult to use, or even unusable, and some thesis authors do not share the source code, which cannot guide AI developers to better reference and explore the current situation, Huawei cloud officially launched some classic and commonly used thesis algorithms on the Huawei cloud one-stop AI development platform modelarts for free use by AI developers, reducing the learning threshold for developers; At the same time, Huawei cloud AI classic paper reproduction activity is launched to encourage participants to reproduce AI classic papers based on modelarts, and open and share the reproduced paper algorithm to modelarts AI Gallery, so as to create an AI developer learning community, so as to achieve the goal of empowering AI developers and allowing more people to use classic algorithms at a low threshold

at present, more than 100 classic AI algorithms have been launched in Huawei cloud AI Gallery, which provides enough strength for the development of zigzag reinforcement. People can understand the algorithm introduction, view the code, and one click subscription algorithm, and conduct training and reasoning on the modelarts platform. Huawei cloud modelarts product team said that by combining these hundreds of AI classic paper algorithms with existing tutorials, a set of AI learning ladders from simple to deep will be built to help developers from all walks of life get better in the field of artificial intelligence

full support for AI classic paper reproduction

in order to enable developers to better learn AI, Huawei cloud organized the AI classic paper reproduction activity, which set up a million level bonus pool. If it is successfully reproduced and passed the acceptance, you can get bonus incentives. Each paper is reproduced by a team. Once the paper is published, it will attract students from major well-known universities to participate. Students can choose familiar papers based on their own research fields. In addition, in order to help students successfully complete the reproduction, the activity provided all-round support to the registered students, including the promotion of modelarts empowerment and paper reproduction activities, providing sufficient modelarts. It is calculated that it uses a diamond cone with a vertex angle of 120 ° or a steel ball with a diameter of 1.59, 3.18mm, full cycle guidance and problem solving. The whole activity was highly praised by the students

representative AI classic paper algorithm: image segmentation fast SCNN

with the rise of autonomous vehicle, there is an urgent need for a model that can deal with the input of downstream market segmentation needs in real time. At present, there are some advanced offline semantic segmentation models, but these models have large volume, large memory and large amount of computation. Fast scnn: fast semantic segmentation network can solve these problems. It is a real-time semantic segmentation model for high-resolution image data, which is suitable for efficient computing on low memory embedded devices

representative AI classic paper algorithm: target detection TSD

at present, many studies show that there is a large deviation between the classification branch and the positioning branch in target detection. Starting with the reconstruction of the sibling head, the paper jumps out of the conventional optimization direction and proposes the tsd:revising the sibling head in object detector method to solve the internal conflict caused by mixed tasks. This method can be flexibly inserted into most detectors, Help the general detector greatly improve the performance by 3%-5%. On coco, the map based on resnet-101 can reach 49.4, and on senet154 can reach 51.2

learn AI with modelarts

according to the Huawei cloud modelarts product team, in 2021, the Huawei cloud AI classic thesis reproduction plan will bring more follow-up activities: first, invite excellent student representatives to hold a live interpretation of thesis reproduction, and lead developers to reproduce classic thesis; 2、 Publish more AI classic papers from time to time; 3、 Invite outstanding student representatives to attend Huawei Developer Conference 2021 (cloud) and have face-to-face exchanges with experts

Copyright © 2011 JIN SHI