BEGIN:VCALENDAR
VERSION:2.0
PRODID:Data::ICal 0.22
BEGIN:VEVENT
DESCRIPTION: '\n\n Machine Learning Model Hardening For Fun and Profit\
n\n Ariel Herbert-Voss\n\n Machine learning has been widely and enthus
iastically applied to a\n variety of problems to great success and is in
creasingly used to\n develop systems that handle sensitive data - despit
e having seen that\n for out-of-the-box applications\, determined advers
aries can extract\n the training data set and other sensitive informatio
n. Suggested\n techniques for improving the privacy and security of thes
e systems\n include differential privacy\, homomorphic encryption\, and
secure\n multi-party computation. In this talk\, we’ll take a look at
the\n modern machine learning pipeline and identify the threat models th
at\n are solved using these techniques. We’ll evaluate the possible co
sts\n to accuracy and time complexity and present practical application
tips\n for model hardening. I will also present some red team tools I\n
developed to easily check black box machine learning APIs for\n vulner
abilities to a variety of mathematical exploits.\n\n Ariel Herbert-Voss
is a PhD student at Harvard University\, where she\n specializes in deep
learning\, cybersecurity\, and mathematical\n optimization. Like many m
achine learning researchers\, she spent plenty\n of time thinking about
deep learning from a computational neuroscience\n point of view without
realizing that skulls make biological neural\n networks a lot less hacka
ble than artificial ones. Now she thinks\n about securing deep learning
algorithms and offensive applications.\n\n '\n\n
DTEND:20180811T202000Z
DTSTART:20180811T200000Z
LOCATION:AIV - Caesars Promenade Level - Florentine BR 3
SUMMARY:Machine Learning Model Hardening For Fun and Profit
END:VEVENT
END:VCALENDAR