GTC-DC 2019: Explainable Deep Learning to Fight Sex Trafficking
Robert Pless, GWU
We highlight recent algorithms and results for explainable AI in the context of deep learning for visual search, motivated by a project called Traffickcam.This tool to fight human trafficking is trained for investigatory searches to find the hotel where a trafficking victim was photographed. It's currently used by the National Center for Missing and Exploited Children and by law enforcement investigators to characterize the type and extent of trafficking operations. We'll share and describe the search tool, which is based on Deep Distance Metric Learning.We'll also share our derivations of new visualization tools inspired by the need to explain to investigators why Deep Distance Metric Learning considers two images to be similar.We'll also describe quantitative results based in our dataset, Hotels-50K, composed of more than 1 million images from 50,000 hotels worldwide,which we share to encourage additional work in this area.