laitimes

CNCC | AIGC Content Chaos: Can Detection and Traceability Dominate the World?

CNCC2024

Brief introduction of the forum:

AIGC Content Chaos: Can Detection and Traceability Dominate the World?

Time: 13:30-17:30 on October 25

Location: Hall 3, 1st Floor, Xia Yuan-Haiyan Hall

Note: If there is any change, please refer to the final information on the official website (https://ccf.org.cn/cncc2024).

The breakthrough progress of AIGC technology has greatly enriched the digital media ecosystem, making creation more convenient and diverse than ever. From generating artwork to writing articles, AIGC has revolutionized content production by providing creators with powerful tools. However, the rapid development of this technology has also brought a series of negative effects, especially the frequent occurrence of fake content scams, which have forced people to face the risks of anonymity and untraceability of online media.

There are still some problems to be solved in the research on AIGC content detection and traceability. Firstly, the existing detection technology is difficult to effectively identify complex and diverse AIGC-generated content, resulting in a low identification rate of forged information. Second, many tools lack real-time visibility to keep track of newly generated content, exacerbating cyber security risks. In addition, the lack of a unified detection framework due to different standards across platforms adds complexity to content governance. Finally, how to ensure transparency and traceability while protecting user privacy remains a key challenge for researchers and policymakers.

This forum focuses on the detection and traceability of AIGC content, and discusses how to accurately and efficiently identify and track AIGC-generated content to reduce security risks. We will discuss in depth how technological innovation can improve the accuracy and efficiency of detection, enhance the ability to identify counterfeit content, and build a comprehensive content traceability system. At the same time, the forum will also discuss how to realize the healthy development and standardized application of AIGC technology under the premise of ensuring the diversity and authenticity of content.

Forum Agenda

CNCC | AIGC Content Chaos: Can Detection and Traceability Dominate the World?

Introduction of the chairman and guests of the forum

Chair of the Forum

CNCC | AIGC Content Chaos: Can Detection and Traceability Dominate the World?

Ren Wenqi

Member of CCF YOCSEF Shenzhen AC, professor of Sun Yat-sen University, excellent youth

Introduction: Executive member of CCF Multimedia Committee, member of CCF YOCSEF Shenzhen AC, introduced talents of Sun Yat-sen University's "Hundred Talents Program", excellent youth, professor, won the Outstanding Doctoral Dissertation Award of China Computer Federation, Wu Wenjun Artificial Intelligence Outstanding Youth Award, First Prize of Natural Science of China Image and Graphics Society, etc.

Co-Chairs

CNCC | AIGC Content Chaos: Can Detection and Traceability Dominate the World?

Qi Shuhan

Academic Secretary of CCF YOCSEF Shenzhen, Associate Professor of Harbin Institute of Technology (Shenzhen).

Introduction: Executive member of CCF Multimedia Committee, academic secretary of CCF YOCSEF Shenzhen, deputy director of Guangdong Provincial Key Laboratory of Decision Intelligence. He has won the second prize of scientific and technological progress of the Ministry of Education, the gold medal of the "Challenge Cup" science and technology innovation competition, and the special prize of the China University Computer Education Conference.

Forum Speaker

CNCC | AIGC Content Chaos: Can Detection and Traceability Dominate the World?

Peng Yuxin

Professor of Peking University, Chief Expert of the 863 Project, Outstanding Youth

Introduction: Boya Distinguished Professor of Peking University, Chief Expert of 863 Project, Jie Qing, Director of the Expert Committee of the Chinese Artificial Intelligence Industry Innovation Alliance, Expert of the Expert Committee of the "Artificial Intelligence 2.0" Planning of the Chinese Academy of Engineering. As the first finisher, he won the first prize of the 2016 Beijing Science and Technology Award and the first prize of the 2020 Science and Technology Progress Award of the Chinese Institute of Electronics.

Title: Understanding and Generating Fine-grained Multimodal Scenarios

Abstract:Scenes are the basic elements of the world, and the accurate understanding and generation of scenes is one of the important applications to promote the implementation of large models. Fine-grained, multi-modal scene understanding and generation can more accurately portray the real world, and can be widely used in smart cities, smart manufacturing, smart retail and other fields. Fine-grained multi-modal scene understanding and generation aims to identify the fine-grained information of multi-modal objects in the scene to achieve scene understanding, and to generate semantically controllable and realistic scene generation, and the key scientific problems include how to accurately identify any category in the open domain, how to continuously add fine-grained category objects, and how to generate multi-modal scenes that meet the requirements according to the fine-grained attributes and relationships of objects. This report will introduce the recent work progress of the research group in related fields, and discuss and look forward to the future development direction.

CNCC | AIGC Content Chaos: Can Detection and Traceability Dominate the World?

Zhang Weiming

Professor of University of Science and Technology of China, Vice Dean of the School of Cyberspace Security

Introduction: Professor of University of Science and Technology of China, Vice Dean of the School of Cyberspace Security, "Chief Technical Expert" of the Key Basic Research Project of the National Foundation Strengthening Program. He has won the first prize of military scientific and technological progress, the first prize of Anhui Provincial Natural Science Award and other awards.

Title: Monitoring and Detection Technology of Large Model Generation Content

Abstract:The rapid development of generative artificial intelligence based on large models has brought convenience to people's learning, work and life, but also brought new challenges to cyberspace security. On the one hand, the content output of the large model will affect human cognitive security, such as false information cognitive interference, human information ecological pollution, illegal abuse of personal data, etc. On the other hand, it is difficult to control the thoughts and thoughts of large model agents, which induces the emerging problem of machine cognitive safety. This report will discuss the impact of large models on machine cognitive safety and human cognitive safety from two perspectives: internal monitoring and external detection.

CNCC | AIGC Content Chaos: Can Detection and Traceability Dominate the World?

Liu An'an

Professor and outstanding youth of Tianjin University

Introduction: Director of the Institute of Graphics of Tianjin University, he won the special prize of Tianjin Science and Technology Progress Award as the first completer, and published more than 100 high-level papers as the first/corresponding author.

Title: Endogenous Risk Analysis and Governance of Visual Large Models

Abstract:There are endogenous risks in visual generation large models (such as Stable Diffusion and Midjourney series models), and users can use prompts without explicit risks to attack the model to generate risk images. A typical solution to this problem is to use a series of risk factor filters to avoid the generation or propagation of risk images. However, due to the unknown mechanism of risk generation generation model, the generation of risk images cannot be blocked from the perspective of "root cause". This report will introduce the team's research on the endogenous risk mechanism of the visual large model, and propose corresponding risk defense strategies. The report first introduces a general attack framework that induces endogenous risks in large visual models. Based on the attack results, we excavate the patterns that induce the model to generate risky content from the text prompt and feature space, respectively, and finally propose a corresponding defense framework based on the risk-induced model to enhance the security of the model-generated content.

CNCC | AIGC Content Chaos: Can Detection and Traceability Dominate the World?

David Wang

Deputy Director of the Science and Technology Development Department of the National Key Laboratory of Communication Content Cognition

Title: Exploration and Practice of AIGC Content Security Management

Abstract:The wide application of generative artificial intelligence has put forward new requirements for the cultivation of network civilization in the new era. The cultivation of network civilization in the era of intelligent Internet should not only pay attention to people's civilized behavior, but also pay attention to the "education and guidance" of artificial intelligence systems and algorithms. "Using AI to govern AI" trains AI systems with good moral judgment and code of conduct to guide, constrain, and improve human civilization in cyberspace, realize the deep integration and co-evolution of machine intelligence and human intelligence, and promote the development of cyber civilization in a healthier and more harmonious direction.

CNCC | AIGC Content Chaos: Can Detection and Traceability Dominate the World?

Li Wenjin

Technical Director of NSFOCUS Security Competence Center and Head of Tianyuan Laboratory

Introduction: Won the second prize of the 2023 Strong Network Cup AI Artificial Intelligence Challenge, and the second prize of the 2023 CCF Science and Technology Achievement Award for Scientific and Technological Progress.

Title: A New Paradigm of Security Defense for LLM Applications

Abstract:With the implementation of LLM on the business side, attack methods such as model jailbreak and character escape can affect and change the behavior of model applications, which not only affect its functions, but also may leak important data and endanger system security. From the perspective of AI Red Team, this report introduces a variety of security risks and attack methods existing in the mainstream application architecture of large models, and proposes a set of LLM Guard detection and defense ideas based on the LLM Guard mechanism, LLM Guard model and LLM Guard Prompt endogenous defense mechanism to deal with these security risks.

About CNCC2024

CNCC2024 will be held on October 24-26 in Hengdian Town, Dongyang City, Zhejiang Province, with the theme of "Developing New Quality Productivity, Computing Leads the Future". The three-day conference included 18 invited reports, 3 conference forums, 138 thematic forums, 34 thematic activities and more than 100 exhibitions. More than 800 speakers, including Turing Award winners, academicians of the Chinese Academy of Sciences and the Chinese Academy of Sciences, top scholars at home and abroad, and well-known entrepreneurs, looked forward to cutting-edge trends and shared their innovative achievements. More than 10,000 people are expected to attend.

Read on