The Fight Against Child Exploitation: Meta's PDQ Algorithm and Faiss Similarity Search

In 2023, the National Center for Missing and Exploited Children (NCMEC) Cyber Tipline received over 36 million reports of suspected child sexual exploitation. This ever-increasing number has been exasperated with the rise in generative AI and concerns have grown alongside the issue. Many companies have emerged with technologies to aid in the detection of online Child Sexual Abuse Material (CSAM) and support content moderation efforts. Amongst those companies is Meta, which has image and video hashing algorithms dedicated to abuse detection through similarity matching and sharing of hashes. The open-source image hashing algorithm, PDQ, is used alongside a Faiss similarity search, where hashes can be gathered for comparison both internally on Meta platforms and externally with other organizations. In this demo, participants will learn about how PDQ and Faiss work by utilizing a general image dataset and performing various image transformations to assess similarity resilience. Not only will participants have a better understanding of the algorithms in play, but they will also be more informed on the magnitude of CSAM proliferation and the significance in reducing child (re)victimization.

Session Presenters

UyenNghi H. Nguyen

Vice President

./Script

Students focus on strengthening their cybersecurity skills with an emphasis on cryptography and public key infrastructure (PKI), reverse engineering, secure coding and malware analysis, security analysis and incident response, and pentesting. Members create new solutions and contribute to existing ones, building on their knowledge and skills with hands-on experience.