Overordnede kursusmål
Differential Privacy is a field that interests the communities in
machine learning, algorithms, statistics, and algorithms. The
purpose of this course is to get an understanding of Differential
Privacy to the level where one can start incorporating methods and
ideas from Differential Privacy in one's own research.
Læringsmål
En studerende, der fuldt ud har opfyldt kursets mål, vil kunne:
- Understand the setup and the motivations for Differential
Privacy
- Prove that a given algorithm is epsilon-differentially private.
We cover worst-case as well as average-case analysis.
- Apply basic techniques to differentially private
algorithms/methods
- Argue about the behaviour of differentially private algorithms
in an online and streaming setting
- Understand and explain some generalisations of the notion of
Differential Privacy
- Understand, prove, and apply lower bounds for differentially
private algorithms
- Apply differential privacy to machine learning problems
- Evaluate and compare differentially private algorithms for
solving a given problem.
Kursusindhold
We go through the book "The Algorithmic Foundations of
Differential Privacy by Cynthia Dwork and Aaron Roth". Content
is the motivation, basic terms, definitions, theorems, composition
theorems and other basic techniques, queries with correlated error,
generalisations, boosting for queries, average-case analysis, lower
bounds, computational complexity, mechanism design, machine
learning, and additional models, of/for/with/by differential
privacy. We put theory into practice with an implementation
exercise.
Litteraturhenvisninger
The Algorithmic Foundations of Differential Privacy by Cynthia
Dwork and Aaron Roth
Sidst opdateret
06. juni, 2019
