DNV GL signs an agreement with Fudan University

Exploring the application of adversarial generative networks in artificial intelligence risk management.

The Artificial Intelligence Research Center (AIRC) of the Technology and Research (GTR) department of DNV GL Group has launched research cooperation with Shanghai Fudan University to expand its academic research network in China.

DNV GL Artificial Intelligence Research Center and Fudan University Big Data School signed a cooperation agreement to explore the application of confrontation generation network (GAN) in artificial intelligence risk management.

"We are very pleased to sign a cooperation agreement with Dr. Fu Yanwei and Professor Xue Xiangyang's research team at the School of Big Data at Fudan University," said Dr. Pierre Sames, R&D Director of DNV GL Group. "This cooperation has greatly enhanced our artificial intelligence capabilities. This is part of our AIRC, the center focuses on the two fields of computer vision and intelligent IoT devices. AIRC's mission is to use artificial intelligence technology to enhance our existing services, and help DNV GL achieve greater goals, for artificial intelligence algorithms Provide risk management."

Chen Zhongning, head of the DNV GL Artificial Intelligence Research Center, added: "As a risk management company with a history of more than 150 years, DNV GL realizes that this new technology of artificial intelligence algorithms also brings unprecedented efficiency improvements. The risk challenge has never been seen before. We are very pleased to cooperate with Fudan University in the field of artificial intelligence risk management, and finally provide a feasible and effective method to deal with the security challenges brought by artificial intelligence."

Generative confrontation network (GAN) is a deep neural network system architecture invented in 2014, in which two neural networks confront each other, similar to competitive games. In a specific training set, this technique generates new data by learning the same statistical data as the training set. Our idea is to generate new images through GAN training on a set of images provided by customers, which not only have realistic features, but also add new detailed features, and are used to check the robustness of deep learning algorithms created by customers Performance, such as the detection and classification of ships. Ultimately, this method will help us build a tool for artificial intelligence algorithm authentication, and provide an effective supplement to the framework introduced in DNVGL-RP-0510 "Data Driven Algorithms and Models Support Framework".

Oriental scholars, young scholars of the "Thousand Talents Program", ARC DECRA Fellow of Australia, and Professor Fu Yanwei of Fudan University's School of Big Data believe that "There is no doubt that deep learning algorithms have achieved remarkable results in many aspects, such as target recognition, Machine translation, speech recognition, etc. The robustness of deep learning algorithms is increasingly important in terms of commercial deployment and application possibilities in safety-critical systems."

"The research on artificial intelligence algorithm certification is extremely important and urgent. For example, criminals may make adversarial traffic signs, causing autonomous vehicles to take unnecessary actions. We are very happy to cooperate with DNV GL to jointly explore the use of artificial intelligence Algorithm-tested GAN." Xue Xiangyang, vice president of the School of Big Data and the Institute of Big Data at Fudan University, said.