Ali Joined Hands With Tsinghua University To Launch "Security AI Challenger Program"

Recently, the fourth phase of "security AI Challenger program" jointly sponsored by Ali security and Tsinghua University and supported by CIKM has been officially launched.

Ali and Tsinghua both put forward higher requirements for the difficulty of the competition. First of all, in terms of competition questions, Ali security and Tsinghua experts jointly created the world's first anti attack scenario combining black and white boxes for multiple target detection models.

In addition, in addition to the participation of university students such as master's degree, it also attracted a large number of social players to participate in the challenge. What's more, Alibaba security also opened a special staff session for the first time to let Ali's technology God parachute to participate in the attack. Up to now, 1238 teams have signed up and more than 1400 competitors have participated.

According to reports, the Ali security AI Challenger competition uses coco data set, which contains 20 types of objects. The task is to make the typical target detection model can't detect the object in the image by adding the counter patch to the original image, so as to bypass the target localization.

To ensure the difficulty of the competition, four recent state of the art detection models were selected as attack targets, including two white box models, Yolo V4 and faster RCNN, and two unknown black box models.

In addition, the Organizing Committee launched the "free server" for the server shortage of the "Challenger". At present, two batches of servers have been released, and the final batch will be announced on August 31.