Data Exchange Podcast (Episode 115): Nicholas Boucher

Описание: Episode Notes: https://thedataexchange.media/imperce...
Nicholas Boucher is a PhD at Cambridge University where his focus is on security including on topics like homomorphic encryption, voting systems, and adversarial machine learning.
**Sections** ↓
Introduction to Nicolas Boucher - 00:00:04
General types of attacks and spam detection - 00:01:46
ML Security and attacks on models  - 00:05:32
Responsible AI and growth of ML as the discipline grows - 00:09:29
What led you to NLP models in particular? - 00:10:25
Real world examples and specific tasks - 00:11:59
What class of application are we talking about?  - 00:12:45
4 types of attacks: Invisible characters, homoglyphs, reorderings, and deletions - 00:13:58
Invisible Character Attacks - 00:14:13
Invisible characters: Do you find characters not seen in training? - 00:18:43
Did you create a testing tools that pointed to an NLP application? - 00:20:37
It is conceivable that we will have tools like this in 2-3 years from now? - 00:21:23
What are some of the other NLP attacks in your paper that scare you? - 00:24:16
Homoglyph Attacks - 00:24:36
Reordering Attacks - 00:28:56
Deletion Attacks - 00:30:43
To what extent should NLP teams be worried about these attacks? - 00:32:04
NLP and verticals where text is important - 00:33:24
Security practitioners and how they think about breaking into a system - 00:38:23
Supply chain problem in all of software - 00:39:35
Homomorphic Encryption and Fully Homomorphic Encryption (FHE) - 00:42:00
Security domain, security of a system, and social engineering - 00:46:21