The billion dollar fraud and how deep learning might avoid it

Banks are being hacked all the time.  According to various statistics, banks get over a million cyberattacks a year, and protecting the bank from breach is getting harder and harder.  This was well illustrated by the Kapersky report of a cybercrime group in February gaining access to over $1 billion in two years by targeting over 100 banks in 30 countries. 

This is just one of many however, as card schemes are being breached almost daily and massive amounts of online card numbers released for very small cost.  Three major breaches released over 300 million card numbers alone (Heartland, TJX and CardSystems Solutions), and you can buy most card or bank details for an average $5 per account.

Source: InformationisBeautiful 

So I claimed the other day that PayPal are the most secure processor in the world, as they are one of the most attacked.  PayPal does get cracks – they were breached at least twice last year, once by a white hat hacker and similarly by a software company - but overall are pretty secure.

What’s PayPal’s secret?

Well obviously, they’re not going to tell you but they do release a few clues.  For example, via the PayPal engineering blogs, we learn that they are using deep learning via Hadoop to build intelligence into the network.  In another hint at what’s to come Hui Wang, PayPal’s Senior Director of Global Risk Sciences, talks about  the idea that their systems will soon be able to retrain and adapt in real-time to deep learning.

In other words, banks, payment processors and other financial firms will soon move to real-time analytics and artificial intelligence techniques to crack down on fraud.

You can see more on PayPal’s deep learning techniques from this presentation by Venkatesh Ramanathan, Data Scientist at Paypal, from December 2014: