Through a Graph, Darkly

*Abstract:*There is now an overwhelming wealth of data sources (e.g. 
smart phones) about human social and mobile activities. These resources 
allow many useful applications: (1) Self-referential uses include: 
network planning for cellular; offloading and opportunistic network 
routing; mobile and opportunistic content distribution. (2) Applications 
include: planning transportation; crowd management; activity-based 
energy monitoring, and disease epidemic tracking. (3) Commercial use 
includes: location based advertising; social and mobile based 
recommendation; social network analysis; urban planning.

Much goodness potentially flows from this work.However, all of this 
comes at a massive cost to personal privacy. To make matters worse, 
there are trends towards data sharing and publication which (quite 
rightly) are placing pressure on researchers to share the evidence for 
their work for reasons of validation, but also (as with medical epidemic 
(and other) data) to allow as yet unforeseen applications to be run on 
the real-world datasets.

Yet historical data about human movements and encounters is intensively 
personal. More so even than mere browsing data, which has already caused 
major furores when inadequately anonymized historical data was released 
(by AOL).Anonymiszing (hyper)graph data is nigh on impossible according 
to current understanding.In this talk, I will attempt to lead a 
discussion about these tensions, and what we might do about this 
practically. There are several challenges including: culture and 
conventions for ethical collection of data, and the evolution of the 
public understanding and informed consent; the technical use of 
differential privacy, and private preserving queries on such data; 
alternative ways of building applications that do not require recording 
such data more than momentarily and locally; and more.