LG CNS 4th in AI image recognition competition

Two panda images are shown in this photo provided by LG CNS, Thursday. The left shows a clear image of panda while the right has noise in the image, which could make it harder for the artificial intelligence (AI) technology to recognize it as panda. But the firm's image recognition technology enables AI to correctly identify objects in blurred photos. / Courtesy of LG CNS

By Baek Byung-yeul

LG CNS finished fourth at an international image recognition competition using artificial intelligence (AI) technology, the LG Group IT affiliate said Thursday.

The event called Adversarial Vision Challenge is hosted by Neural Information Processing Systems (NeurIPS), one of the world's top AI conferences. It is to evaluate how vulnerable AI-computer vision is to adversarial attacks ― designed to fool or decrease the accuracy of the AI model.

LG CNS said being placed fourth proves the superiority of its AI image recognition technology, which can be used in autonomous vehicles.

“The competition is to evaluate how well the AI vision models of participants are working. For instance, an image of panda that looks normal to a human can contain noise that makes an AI system think it is looking at an image of a gibbon,” an LG CNS official said.

The competition, which was conducted for three months starting late July, featured 340 teams and an LG CNS team was the only team from a corporate organization among top five teams. A team from Carnegie Mellon University won the competition, followed by Tsinghua University and Ecole de Technologie Superieure in Montreal, Canada.

Teams participating in the competition were able to take on the role of attacker, defender or both. The defense team is required to build a visual object classifier that is as robust to image perturbations as possible while the attacker is supposed to find the smallest possible image perturbations that will fool the classifier. LG CNS took the defending role.