Speaker: Tswelopele, BSc(Hons) Math UCT | MWR CyberSec
Abstract
This talk will go through some ideas behind my research proposal for my Masters. The proposal puts forth a contribution towards a systematisation of privacy guarantees for machine-learning model-inference with explicit assumptions about malicious agents and trust settings. The hypothesis is that the field of Quantitative Information Flow (QIF) provides a framework for modelling such privacy guarantees and, even more interestingly perhaps, we postulate that all relevant actions taken by a malicious agent against a machine learning model can be seen as an agent performing a POMDP.
In this talk, we will give a brief introduction to the field of QIF with an example, generalise this example to a machine learning context, and then use the generalised example to describe a mapping between concepts in QIF to a POMDP.