The final review and competition of the 2024 Unveiling the List and Taking up the Challenge special event of the 19th Challenge Cup National College Students' Extracurricular Academic Science and Technology Competition was successfully held in Zhejiang from November 15th to 17th. An academic team from State Key Laboratory of Blockchain and Data Security competed with 75 teams from over 30 universities under the topic Research on Multi-Modal Risk Content Recognition Technology Based on Large Models, ultimately winning the national grand prize, fully demonstrating the team's outstanding research literacy and innovative practical capabilities.
Competition Introduction
The Challenge Cup National College Students' Extracurricular Academic Science and Technology Competition is co-hosted by the Central Committee of the Communist Youth League, the China Association for Science and Technology, the Ministry of Education, the Chinese Academy of Social Sciences, the National Student Federation, and the provincial government of the host location. It is praised as the Olympic event of contemporary college students' scientific and technological innovation. With the theme You Set the Challenge, I Take up the Gauntlet, the special competition invites enterprises and institutions to propose demands, and the organizing committee issues a call to heroes to universities and society. This year's Unveiling the List and Taking up the Challenge special competition attracted more than 27,000 participants, with over 6,500 entries. Ultimately, more than 1,950 works from over 410 universities and enterprises across the country were shortlisted for the final review and competition.
Topic Introduction
The unveiled challenge topic participated by the team, Research on Multi-Modal Risk Content Recognition Technology Based on Large Models, was issued by China Unicom, focusing on new challenges in social management and cyberspace governance. In response to multi-modal risk content such as fake faces, scam calls, phishing emails in the era of generative AI, it calls for utilizing the powerful processing capabilities, deep semantic understanding, complex pattern recognition, and continuous evolution capabilities of large models to study and develop more effective multi-modal risk content recognition technologies.
Project Introduction
The team's project, Deep Synthesis Detection and Sensitive Content Management Platform, flexibly applies large models to govern other large models, proposing six technologies including cross-domain knowledge transfer, adaptive sparse activation, multi-modal semantic conversion, and complex thinking chain guidance. It breaks through the generalization and credibility technical difficulties in governing sensitive and illegal content, effectively utilizing and guiding the complex pattern recognition abilities, deep semantic understanding, and model continuous evolution capabilities of multi-modal large models. The project addresses three major challenges: difficulty in generalizing fake content detection, precision in reviewing diverse sensitive types, and dynamic evolution of risk content leading to lagging reviews. It builds a highly accurate, capable, and user-friendly multi-modal risk content review platform.
Introduction to the Team from the National Key Laboratory of Blockchain and Data Security
The core members of the winning team come from the Artificial Intelligence Security Team led by Professor Ren Kui, Deputy Director of the National Key Laboratory of Blockchain and Data Security. The team has long focused on deep synthesis detection and AIGC security, actively responding to the national call. Relying on the National Key Laboratory of Blockchain and Data Security, they have conducted key technology breakthroughs around deep synthesis detection and AIGC security, developing platforms for image deep synthesis detection, deep synthetic voice detection, multi-modal sensitive content review, etc. Through autonomous construction and systematic integration, they have established a multimillion-level multi-modal risk content data foundation and hundreds of content synthesis and detection methods. Their aim is to enhance the capability of scenario-based regulatory compliance for relevant laws and regulations and promote the safe and stable development of generative AI. The supporting technologies of this competition entry have been deployed in several of the aforementioned platforms.
The team from the State Key Laboratory of Blockchain and Data Security won the grand prize at the 19th Challenge Cup 2024 Unveiling the List and Taking up the Challenge special competition.