"Science and Technology Innovation Board Daily" reported on September 7 (Special Correspondent Chen Junqing) , on the afternoon of September 6, the Bund Conference AI Innovation Competition·Global Deepfake Offensive and Defense Challenge kicked off.
, the representative of the top players who uses AI technology to fight against deepfake, appeared in anticipation of the whole network. Mr. Zhu, an audience member from Shanghai, said that it was the first time he saw the head-to-head confrontation between generative AI and deepfake detection technology with the naked eye. The competition was so interesting.
▌artificial pk artificial intelligence, which one is better?
After a brief award ceremony for the top players, the competition between the players started again.
The School of Cyberspace Security of the University of Science and Technology of China and Ant Digital Tianji Laboratory jointly developed the question to conduct the ultimate examination of the player models. Like the online competition, the exhibition competition is divided into an image track and an audio and video track, and the top three players in each track compete against each other in pairs. As soon as the
image track started, the audience immediately became agitated. Host Bi Dao summoned 9 photos of Marilyn Monroe, 2 of which were forged by AIGC. There are 9 pictures that are difficult to distinguish between authenticity and fakeness with the naked eye, but contestants need to use their own competition models to give the probability of forgery for each picture, and compare the two pictures with the highest probability values with the correct answers. "I thought I would just watch the fire from the other side, but I never thought that I would be involved in a battle, trying to compete with AI with my naked eyes," said Xiao Zhang, an audience member from a technology company. During the audience interaction session, the unconvinced visual designers in the audience even asked AI to fight, hoping to show their professional sensitivity. The problem of the probability value of
prompted Director Bi to ask whether AI can 100% identify forgery. Yao Weibin, one of the question makers of the
competition and technical director of zoloz, said that it is very difficult for AI to 100% identify whether a forged image is fake. Counterfeiting technology is also constantly iteratively updated, and confrontation needs to be continuously improved. This is also a major significance of our competition today. In practical applications, identification algorithms are usually used in combination with other detection methods to comprehensively judge risks.
It is reported that the deepfake competition covers the most mainstream picture and image tracks, covering the entire process from basic scientific research to commercial transformation. The data set consists of public data and forged data. Among them, the forged picture data covers more than 50 generation methods in real scenes, and the fake audio and video includes more than 100 combination attack methods. The total training data set exceeds 1 million. In the
video track, Bi Dao’s digital avatars starred in many famous scenes from classic movies, and contestants also had to use AI models to identify authenticity. In terms of on-site contestant scores, on some moderately difficult questions, the results were close, and the forgery recognition rates were almost all above 80%. However, on some highly difficult questions, the model results were very different.
Bi Dao directed a soul torture to Professor Tianyi from the Singapore Agency for Science, Technology and Research. Why does the AI can't completely answer a picture that looks fake to the naked eye? And the picture that is difficult to distinguish with the naked eye can be recognized by AI in 1 second?
In this regard, Zhou Tianyi believes that humans rely on intuition and logic to judge whether an image is fake. For example, when Director Bi’s face was replaced with Harry Potter’s, humans can tell at a glance, but AI does not understand the story behind it. While humans will be lazy subconsciously, AI is the most competent tool man. AI is not disturbed by emotions and will complete human instructions meticulously and diligently. Therefore, AI is much higher than humans in terms of work efficiency and stability.
▌ "AI face-changing" detection has a standard
Recently, the deepfake incident has affected people's hearts.
On August 30, the competition organizing committee launched an open source initiative: "Support and encourage outstanding participants to open source competition models, lower the technical threshold, strengthen technical exchanges, and then help more people detect counterfeit content and help AI do good.
develops detection algorithms , is just one of the ways to curb the abuse of deepfake technology. Yao Weibin admitted that this method is relatively lagging behind. If fraud is to be prevented from the root, Yao Weibin believes that a universal label should be established for AIGC content from a standard and legislative perspective. Each video/picture generated by AIGC will have a signature or watermark in the file information, which will provide more basis for platform management and make AIGC more credible.
At the conference, more than ten institutions jointly released the country’s first “AI face-changing” detection standard for financial scenarios. The release of this standard provides a basis for the security detection and assessment of fake digital faces in financial scenarios, and also fills the gap in this field. The
deepfake attack and defense challenge applies this standard framework and indicators to formulate competition questions and evaluate answers. The
Global Deepfake Offensive and Defense Challenge conducts practical offensive and defensive drills on the fraud risks of "AI face-changing", and sets up a prize pool of RMB 1 million to encourage technical talents who promote AI for good. The competition attracted more than 2,200 players and more than 1,500 teams from around the world, covering 26 countries and regions including China, the United States, India, Australia, Japan, Indonesia, Malaysia, Singapore, and Vietnam.
▌Aimed at talent cultivation
It is understood that the Bund Conference AI Innovation Competition includes two major sub-events: "afac2024 Financial Intelligence Innovation Competition" and "Global Deepfake Attack and Defense Challenge". It comprehensively demonstrated the latest applications and cutting-edge exploration of artificial intelligence in multiple fields, and also demonstrated the infinite vitality of scientific and technological innovation. At the same time, the second AI Innovation Competition of the Inclusion Bund Conference "Global AI Offensive and Defense Challenge" was also officially launched. The event focuses on the industrial practice of AI large models and has an offensive and defensive two-way track.
The competition has three major sections: algorithm challenge, application scenario design, and entrepreneurial project roadshow. The algorithm challenge focuses on cutting-edge technologies such as machine learning and deep learning; the application scenario design encourages contestants to apply AI technology to medical care, education, environmental protection, etc. fields; entrepreneurial project roadshows provide a platform for start-ups to showcase their innovative ideas and technical capabilities. Among the
deepfake participating teams, there is a "mixed" team, where a senior leads a group of juniors to "fight monsters and upgrade". The core members of "Team Name" come from universities in Hong Kong and Macau. Wu Haiwei, who just graduated from the University of Macau with a Ph.D. in August last year and is currently a postdoctoral fellow at the City University of Hong Kong, brought his juniors from the University of Macau's Smart City Internet of Things Laboratory. Join the competition together.
As early as 2019, Wu Haiwei participated in the deepfake detection competition organized by Facebook. From 2020 to 2022, Wu Haiwei participated in many related competitions and won the top three places. Tang Yongwei is the most special person in this competition. He graduated from commerce and taught himself computer expertise. This time, we achieved top three rankings in both events.
The name of Tang Yongwei’s team is “The First Pattern of Rain with Xiao Sa”, which comes from Ouyang Xiu’s “Ode to the Autumn Wind”. This ancient name is completely different from his competition style of wearing a scarf and cutting off thorns all the way. At the end of 2022, he participated in a professional competition for the first time and won the third place in the a-tech competition.
Compared with the final results, Yan Qiang, who participated in the afca competition, valued the value of major competitions in talent cultivation and motivation. By participating in the competition, he could keep abreast of industry trends and technology trends, which not only made him stronger, but also made him more powerful. A group of people becomes stronger. Lanman, one of the judges of the
competition and a professor at the School of Computer Science and Technology of East China Normal University, pointed out that the questions in the afac competition are both forward-looking and practical. Some involve basic technical research in the financial industry, and some are related to specific application scenarios. The combination reflects the original intention of the competition to "solve the real proposition of the industry". "Promoting research through competition and promoting research through production, the relevant results are very valuable." The two competitions of
not only reflected the internationalization of the event, but also demonstrated its authority in the professional field. The participating projects cover many hot areas such as medical health, financial technology, and intelligent manufacturing, demonstrating the broad application potential of AI technology.
"Such competitions are helpful to players at different stages." Ma Qianli, an algorithm engineer at Shopee, participated in the AFCA competition for the second time. When he participated in the competition last year, he was still a fresh graduate from college. “When applying for a job, innovation and practical ability are important points to widen the gap with competitors. The afac competition has become evidence of this innovative and practical ability. The competition experience gave me a lot of experience during the interview. A small boost.”
"Science and Technology Innovation Board Daily" reported on September 7 (Special Correspondent Chen Junqing) , on the afternoon of September 6, the Bund Conference AI Innovation Competition·Global Deepfake Offensive and Defense Challenge kicked off.
, the representative of the top players who uses AI technology to fight against deepfake, appeared in anticipation of the whole network. Mr. Zhu, an audience member from Shanghai, said that it was the first time he saw the head-to-head confrontation between generative AI and deepfake detection technology with the naked eye. The competition was so interesting.
▌artificial pk artificial intelligence, which one is better?
After a brief award ceremony for the top players, the competition between the players started again.
The School of Cyberspace Security of the University of Science and Technology of China and Ant Digital Tianji Laboratory jointly developed the question to conduct the ultimate examination of the player models. Like the online competition, the exhibition competition is divided into an image track and an audio and video track, and the top three players in each track compete against each other in pairs. As soon as the
image track started, the audience immediately became agitated. Host Bi Dao summoned 9 photos of Marilyn Monroe, 2 of which were forged by AIGC. There are 9 pictures that are difficult to distinguish between authenticity and fakeness with the naked eye, but contestants need to use their own competition models to give the probability of forgery for each picture, and compare the two pictures with the highest probability values with the correct answers. "I thought I would just watch the fire from the other side, but I never thought that I would be involved in a battle, trying to compete with AI with my naked eyes," said Xiao Zhang, an audience member from a technology company. During the audience interaction session, the unconvinced visual designers in the audience even asked AI to fight, hoping to show their professional sensitivity. The problem of the probability value of
prompted Director Bi to ask whether AI can 100% identify forgery. Yao Weibin, one of the question makers of the
competition and technical director of zoloz, said that it is very difficult for AI to 100% identify whether a forged image is fake. Counterfeiting technology is also constantly iteratively updated, and confrontation needs to be continuously improved. This is also a major significance of our competition today. In practical applications, identification algorithms are usually used in combination with other detection methods to comprehensively judge risks.
It is reported that the deepfake competition covers the most mainstream picture and image tracks, covering the entire process from basic scientific research to commercial transformation. The data set consists of public data and forged data. Among them, the forged picture data covers more than 50 generation methods in real scenes, and the fake audio and video includes more than 100 combination attack methods. The total training data set exceeds 1 million. In the
video track, Bi Dao’s digital avatars starred in many famous scenes from classic movies, and contestants also had to use AI models to identify authenticity. In terms of on-site contestant scores, on some moderately difficult questions, the results were close, and the forgery recognition rates were almost all above 80%. However, on some highly difficult questions, the model results were very different.
Bi Dao directed a soul torture to Professor Tianyi from the Singapore Agency for Science, Technology and Research. Why does the AI can't completely answer a picture that looks fake to the naked eye? And the picture that is difficult to distinguish with the naked eye can be recognized by AI in 1 second?
In this regard, Zhou Tianyi believes that humans rely on intuition and logic to judge whether an image is fake. For example, when Director Bi’s face was replaced with Harry Potter’s, humans can tell at a glance, but AI does not understand the story behind it. While humans will be lazy subconsciously, AI is the most competent tool man. AI is not disturbed by emotions and will complete human instructions meticulously and diligently. Therefore, AI is much higher than humans in terms of work efficiency and stability.
▌ "AI face-changing" detection has a standard
Recently, the deepfake incident has affected people's hearts.
On August 30, the competition organizing committee launched an open source initiative: "Support and encourage outstanding participants to open source competition models, lower the technical threshold, strengthen technical exchanges, and then help more people detect counterfeit content and help AI do good.
develops detection algorithms , is just one of the ways to curb the abuse of deepfake technology. Yao Weibin admitted that this method is relatively lagging behind. If fraud is to be prevented from the root, Yao Weibin believes that a universal label should be established for AIGC content from a standard and legislative perspective. Each video/picture generated by AIGC will have a signature or watermark in the file information, which will provide more basis for platform management and make AIGC more credible.
At the conference, more than ten institutions jointly released the country’s first “AI face-changing” detection standard for financial scenarios. The release of this standard provides a basis for the security detection and assessment of fake digital faces in financial scenarios, and also fills the gap in this field. The
deepfake attack and defense challenge applies this standard framework and indicators to formulate competition questions and evaluate answers. The
Global Deepfake Offensive and Defense Challenge conducts practical offensive and defensive drills on the fraud risks of "AI face-changing", and sets up a prize pool of RMB 1 million to encourage technical talents who promote AI for good. The competition attracted more than 2,200 players and more than 1,500 teams from around the world, covering 26 countries and regions including China, the United States, India, Australia, Japan, Indonesia, Malaysia, Singapore, and Vietnam.
▌Aimed at talent cultivation
It is understood that the Bund Conference AI Innovation Competition includes two major sub-events: "afac2024 Financial Intelligence Innovation Competition" and "Global Deepfake Attack and Defense Challenge". It comprehensively demonstrated the latest applications and cutting-edge exploration of artificial intelligence in multiple fields, and also demonstrated the infinite vitality of scientific and technological innovation. At the same time, the second AI Innovation Competition of the Inclusion Bund Conference "Global AI Offensive and Defense Challenge" was also officially launched. The event focuses on the industrial practice of AI large models and has an offensive and defensive two-way track.
The competition has three major sections: algorithm challenge, application scenario design, and entrepreneurial project roadshow. The algorithm challenge focuses on cutting-edge technologies such as machine learning and deep learning; the application scenario design encourages contestants to apply AI technology to medical care, education, environmental protection, etc. fields; entrepreneurial project roadshows provide a platform for start-ups to showcase their innovative ideas and technical capabilities. Among the
deepfake participating teams, there is a "mixed" team, where a senior leads a group of juniors to "fight monsters and upgrade". The core members of "Team Name" come from universities in Hong Kong and Macau. Wu Haiwei, who just graduated from the University of Macau with a Ph.D. in August last year and is currently a postdoctoral fellow at the City University of Hong Kong, brought his juniors from the University of Macau's Smart City Internet of Things Laboratory. Join the competition together.
As early as 2019, Wu Haiwei participated in the deepfake detection competition organized by Facebook. From 2020 to 2022, Wu Haiwei participated in many related competitions and won the top three places. Tang Yongwei is the most special person in this competition. He graduated from commerce and taught himself computer expertise. This time, we achieved top three rankings in both events.
The name of Tang Yongwei’s team is “The First Pattern of Rain with Xiao Sa”, which comes from Ouyang Xiu’s “Ode to the Autumn Wind”. This ancient name is completely different from his competition style of wearing a scarf and cutting off thorns all the way. At the end of 2022, he participated in a professional competition for the first time and won the third place in the a-tech competition.
Compared with the final results, Yan Qiang, who participated in the afca competition, valued the value of major competitions in talent cultivation and motivation. By participating in the competition, he could keep abreast of industry trends and technology trends, which not only made him stronger, but also made him more powerful. A group of people becomes stronger. Lanman, one of the judges of the
competition and a professor at the School of Computer Science and Technology of East China Normal University, pointed out that the questions in the afac competition are both forward-looking and practical. Some involve basic technical research in the financial industry, and some are related to specific application scenarios. The combination reflects the original intention of the competition to "solve the real proposition of the industry". "Promoting research through competition and promoting research through production, the relevant results are very valuable." The two competitions of
not only reflected the internationalization of the event, but also demonstrated its authority in the professional field. The participating projects cover many hot areas such as medical health, financial technology, and intelligent manufacturing, demonstrating the broad application potential of AI technology.
"Such competitions are helpful to players at different stages." Ma Qianli, an algorithm engineer at Shopee, participated in the AFCA competition for the second time. When he participated in the competition last year, he was still a fresh graduate from college. “When applying for a job, innovation and practical ability are important points to widen the gap with competitors. The afac competition has become evidence of this innovative and practical ability. The competition experience gave me a lot of experience during the interview. A small boost.”
-
-
-
-
-
-
-
-
-
-