Alex Edmonds
I completed my PhD at the University of Toronto, supervised by Aleksandar Nikolov and Toniann Pitassi. The focus of my research was on differential privacy, particularly the local model, with connections to learning theory, information theory, and duality.
Work with Nikolov and Ullman, ‘The Power of Factorization Mechanisms in Local and Central Differential Privacy’ (STOC 2020), gave a general characterization for the sample complexity of answering linear queries in the local model of differential privacy.
Building on the prior work’s implications
for agnostic learning and refutation,
work with Nikolov and Pitassi,
‘Learning versus Refutation in Noninteractive Local Differential Privacy’
(NeurIPS 2022),
showed the equivalence of agnostic learning and agnostic refutaion in terms of sample complexity.
We also demonstrate that realizable refutability implies realizable learnability.
In addition to giving a unified presentation of the previous results, my PhD thesis presents further joint research with Nikolov and Pitassi demonstrating relationships between local differential privacy and other data-access models, including single-intrusion pan-privacy and sequentially interactive local differential privacy.
My MSc thesis, supervised by Dan Roy, explores formalizations of computationally-efficient samplability and their relative complexity.
My undergrad was in the Mathematics Specialist program at the University of Toronto.
Contact

Research
- Alexander Edmonds, Sample-Complexity Optimality Under Local Differential Privacy and Related Models. PhD Thesis, 2023. [thesis]
- Alexander Edmonds, Aleksandar Nikolov, Toniann Pitassi, Learning versus Refutation in Noninteractive Local Differential Privacy, NeurIPS 2022. [arXiv]
- Alexander Edmonds, Aleksandar Nikolov, Jonathan Ullman, The Power of Factorization Mechanisms in Local and Central Differential Privacy, STOC 2020. [arXiv] [short talk] [long talk]
- Alexander Edmonds, Concepts of Efficient Samplability. MSc Thesis, 2017. [thesis]