Presented at
Kiwicon 9: Cyberwar Is Hell (2015),
Dec. 11, 2015, 11:30 a.m.
(30 minutes)
The only way for mass surveillance to scale is through automated signal detection, indexing and annotation of raw surveillance data. Computer vision and machine learning algorithms today can do close to real-time face detection on a raspberry pi. As such, these techniques can economically be applied at scale for mass surveillance of public video feeds.
In our dystopic future of automatically flagging individuals of interest, such automated signal detection seems likely to become more common rather than less (at least in jurisdictions that devalue privacy relative to "national and economic security"). This talk will explore how face detection and recognition algorithms work, demonstrate what failure modes exist for common techniques, and how to exploit them.
Presenters:
-
ferrouswheel
ferrouswheel is a person of many hats. He started off building large-scale ecological simulations for his PhD, then got drawn into helping build a artificial general intelligence framework called OpenCog, this was later integrated with unscripted, next-generation AI computer game characters in Hong Kong.
More recently there was some entrepreneurial rah rah activity, returning to New Zealand, and building giant spinning death traps for the Wellington Lux light festival. Also some backend software development to pay the bills.
Links:
Similar Presentations: