The Global Deepfake Attack and Defense Challenge, an innovative AI competition at the Bund AI Innovation Conference, has officially kicked off with nearly a hundred participants contributing to an open-source model code relay. The event, which aims to use AI technology to combat Deepfake, has drawn widespread attention from the public.
The competition, organized by the School of Cybersecurity of the University of Science and Technology of China and Ant Group’s Tianji Lab, features a diverse range of participants from around the world. The event was held on September 6th, with temperatures soaring to 40 degrees Celsius outside and a fiery passion inside the venue.
Among the participants, one notable figure was the tech blogger Bi Dao, who was amazed to see a digital version of himself, Bi Dao, seemingly stealing his livelihood. The competition showcased the confrontation between generative AI and Deepfake detection technology, providing an exciting glimpse into the future of AI.
The competition was divided into two tracks: the image track and the audio-visual track. In the image track, participants were tasked with identifying forged images among a set of nine photos of Marilyn Monroe, including two AIGC-generated ones. The participants had to use their own models to provide the forged probability values for each image and compare them with the correct answers.
The competition was fascinating to watch, as it showcased the power of AI in detecting forged content, said Zhu, a Shanghai-based viewer. It was intriguing to see how the participants used their models to differentiate between real and forged images.
The audio-visual track featured a digital version of Bi Dao appearing in various classic movie scenes, and participants had to use their AI models to determine the authenticity of the scenes.
In the competition, the scores were close in many moderate-level questions, with the forged recognition rate reaching over 80%. However, the differences in the model results were significant in some high-level questions.
The event also highlighted the importance of collaboration and the open-source community in combating Deepfake. The competition组委会 issued an open-source initiative, encouraging participants to release their models to lower technical barriers and promote knowledge exchange. To date, nearly a hundred participants and teams have agreed to participate in the code relay, aiming to protect potential victims of Deepfake technology.
The event also addressed the need for standards and legislation to prevent fraudulent activities.姚伟斌, ZOLOZ’s Technical Director, suggested that a universal identifier, such as a signature or watermark, should be established for AIGC content to make it more trustworthy and easier to manage.
The Global Deepfake Attack and Defense Challenge has attracted over 2,200 participants from 26 countries and regions, including China, the United States, India, Australia, Japan, Indonesia, Malaysia, Singapore, Hong Kong, and Vietnam. The event is a testament to the power of AI technology and its potential to shape the future.
The Bund AI Innovation Conference, which includes the AFAC2024 Financial Intelligent Innovation Competition and the Global Deepfake Attack and Defense Challenge, showcases the latest applications and cutting-edge research in AI technology across various fields. The event also highlights the limitless vitality of technological innovation.
In conclusion, the Global Deepfake Attack and Defense Challenge is a significant step towards combating the threat of Deepfake technology. The event has demonstrated the power of AI in detecting and combating Deepfake, and has also emphasized the importance of collaboration and open-source communities in this effort.
Views: 0