What To Do When Your Machine Learning Gets Attacked

Vishwanath Ramarao (Impermium)
Data Science Ballroom AB
Average rating: **...
(2.60, 5 ratings)

More people are on social networks like Twitter, Facebook, Tumblr, and Pinterest, or using social elements – like comments – of web sites. With rising participation and changes in the ways people communicate, comes increased risk and challenge.

Spam and security problems have gotten significantly more complicated since Paul Graham’s seminal “A Plan for Spam.” The emergence of unguarded social media channels, easy to operate botnets, and cheap labor that can very quickly produce new strains of an attack have rapidly altered the landscape of problems that today’s systems have to deal with.

Machine learning is a proven approach to solving classification problems, and is often cited as a great fit to solving spam and security challenges as well. However, most spam and security solutions that incorporate machine learning are unable to keep up with the rapidly changing nature of the abuse domain and wear down quickly in effectiveness. As a result, the field is rife with failed machine learning projects. Considering the complexity of a machine learning implementation and the margins for error involved, one has to wonder, does machine learning stand a chance as a long-term approach to abuse and security problems?

Machine learning, when an active adversary is involved, requires a novel approach to feature engineering, training and classification. A simple set of design patterns surrounding the building blocks of machine learning and emerging big data frameworks like Hadoop hold the key to a more secure, abuse free world of communication. This session will discuss how to apply machine learning to create an effective, long-term solution to solving the issue of online abuse and security.

Photo of Vishwanath Ramarao

Vishwanath Ramarao

Impermium

Before joining Impermium, co-founder and CTO Vish led engineering for Yahoo! Mail, Yahoo! Search and Insights groups. Vish has a background in machine learning and optimization but has dabbled in graphics and drug discovery.

Sponsors

Sponsorship Opportunities

For information on exhibition and sponsorship opportunities at the conference, contact Susan Stewart at sstewart@oreilly.com

Media Partner Opportunities

For information on trade opportunities with O'Reilly conferences contact Kathy Yu at mediapartners
@oreilly.com

Press and Media

For media-related inquiries, contact Maureen Jennings at maureen@oreilly.com

Contact Us

View a complete list of Strata contacts