Machine Learning Can Help Find Kids Faster

By May 18, 2017 Deterrence No Comments

Cristian Canton Ferrer is an Applied Research Scientist at Facebook. Cristian joined us this month at Facebook Global Security HQ for a two-day Child Safety Hackathon to help develop, research, build upon and create new solutions in the fight against child sexual exploitation.

Impact, Impact, Impact!

That is the motto many companies infuse into their mission, and their vision is to transform their customer’s life. “Impact” is a difficult-to-quantize entity, by nature highly dimensional and subjective. Quite often, everyone has impact, given the correct spin of the concept.

In one way, impact is a matter of area: the number of people you influence times the value of this contribution. You may have a great publication in a very narrow scientific domain with an incredible number of citations (tall, thin impact), or you may have developed a feature within a product that simplifies a tedious job by a tiny fraction for all users (short, wide impact). Both may have a similar overall area. Both yield to the same absolute impact under this criterion.
I would like to share a bit about a very thin but very tall case: one with the potential to change the life of a single individual forever.

A kid is missing. In the U.S., everyone has seen the flyers depicting the face of someone whose whereabouts are unknown. The National Center for Missing and Exploited Children (NCMEC), is the organization behind most of these efforts. In collaboration with law enforcement, the two aim to shine a light on some of these more desperate cases in hopes of finding the missing child more quickly. There are a multitude of scenarios in which a child may go missing. Unfortunately, there is one that’s quite common: the case of a young run-away where the youth ends up being exploited. Of the more than 11,800 endangered runaways reported to NCMEC in 2015, one in five were likely victims of sex trafficking. No doubt this is a horrific problem that needs to be addressed. This is where technology comes in.

Technology can lend a much-needed hand.

Once a year, Thorn joins efforts with Facebook to organize the Child Safety Hackathon: a 48 hour collaborative event where engineers from all over the tech industry (Google, Microsoft, Amazon, Twitter, Pinterest, Intel, Facebook, and more) work to propose solutions to some of the problems Thorn is tackling.

I was honored to be a part of this event for the second time, serving as the lead of a team of volunteers from several companies. Thus far, this has been one of the most rewarding experiences of my professional career. I have come to realize that if you spend some time figuring out how to pivot your daily work / research into a scenario that may make this world a better place, you may be surprised at the impact your contributions can have.

At this year’s hackathon, our team leveraged some of the latest machine learning libraries open sourced by Facebook to create a prototype of Thorn’s Child Finder Service that could run on a mobile device. Through the prototype, pictures of exploited children were matched against the country’s database of missing children maintained by NCMEC. With this type of technology the reaction time on the part of law enforcement, a crucial factor to help an exploited child, could potentially shrink from days to seconds.

Machine learning has come to the forefront in recent years, enabling hard tasks to become less complex with an the ease of sifting through data and extracting patterns. However, there are many situations in which machine learning can do even more — becoming a key asset in doing good. For me, applying my deep-learning expertise to find missing children brought a different, and very gratifying, dimension to my work.

There is no doubt that among all my professional achievements, contributing to the safeguarding of a single missing and exploited child is the most impactful.
People affected = 1
Magnitude of change = enormous
Lasting impact of the change = forever