Dec 13 2023

Thorn’s Head of Data Science discuss how machine learning can support child safety on content-hosting platforms

Post By: Safer / 2 min read

Product Updates

At the end of November, Thorn’s Head of Data Science, Dr. Rebecca Portnoff, took to the main stage at the 2023 AWS re:Invent conference in Las Vegas. Dr. Portnoff joined AWS CTO Dr. Werner Vogels for his keynote to highlight Thorn's pivotal role in using technology to combat child sexual abuse. Vogels keynote focused on cost-aware architecture and explored how artificial intelligence is impacting our present and our future.

See Thorn’s Dr. Rebecca Portnoff on the Keynote Stage at AWS re:Invent

Dr. Vogels’ keynote is always a big draw at re:Invent and this year was no different. There wasn’t an empty seat in the room when Dr. Portnoff took the stage . Her talk centered on Thorn’s mission to defend children from sexual abuse using technology and specifically focused on how her Data Science team partnered with the Product and Build teams to develop a machine learning classification model (a.k.a. a classifier) to detect child sexual abuse material (CSAM) at a production scale.

During her keynote, Dr. Portnoff explained how Safer’s proactive CSAM scanning technology helps to make a real-world impact for children in active abuse situations.

Dr. Portnoff also shared how, to develop our CSAM Classifier, Thorn’s Data Science team had to turn research on using convolutional neural networks into a classifier that could work at a production scale. Her team followed the Cross Industry Standard Process Data Mining (CRISP-DM) process, a methodology for data mining that follows an established six phase approach.

Diagram illustrating the Cross Industry Standard Process Data Mining (CRISP-DM) process, a methodology for data mining that follows an established six phase approach

A blocker in the development of the Classifier was the fact that sexual abuse content is illegal to possess or host. Thorn’s team first needed to solve for this by establishing relationships and investing in on-premises infrastructure at organizations with the legal right to house this data. With infrastructure in place, her team was ready to prep the data, train, iterate and deploy the Classifier. Regular retraining is part of an ongoing process because as Dr. Portnoff puts it: “Models get stale.”

The impact of the CSAM Classifier for our customers and the potential impact it represents for all content-hosting platforms is made clear when you look at the results. In 2022, Safer customers using the CSAM Classifier flagged 304,466 images and 15,238 videos as potential CSAM.

Thorn + AWS: A critical collaboration

A key piece of the presentation was an emphasis on the collaboration between Thorn and AWS – and how Thorn’s close partnership with AWS and use of its cloud services are vital in scaling up solutions to enable the use of AI technology for the greater social good.

AWS Marketplace also helps to streamline procurement of both the on-prem and API versions of Safer, so that every platform can get these critical tools into the hands of their trust and safety and product teams.

You've successfully subscribed to Safer: Building the internet we deserve.!