I am a Doctoral Researcher in the Information Security Group at University College London, advised by Prof. George Danezis and Prof. Emiliano De Cristofaro.
My research focuses on Machine Learning Robustness and AI Safety, while I also pursue projects on Distributed Systems, Trusted Hardware, Privacy-preserving Statistics and Market Manipulation.
I am currently working on worst-case robustness in ML systems and guarantees that a given trained model fails catastrophically with negligible probability. Moreover, I am experimenting with minimizing side-effects and misspecification in reinforcement learning models.
Besides these, I also work on a variety of other topics usually related with security, privacy, and systems design. Two recent works are Snappy (i.e., a fast payment solution designed for slow permisionless blockchains) and my study on stock exchange infrastructure exploitation (see Publications). Additionally, in cooperation with industrial partners, we released a prototype of our novel high-assurance cryptographic hardware architecture. Moreover, our study on ultrasound tracking received wide-spread attention and is considered the seminal work in this area. With Petr Svenda, we created and maintained the first open-source cryptographic library for JavaCards.
In the past, I developed auditing tools for the Public Key Infrastructure of Deutsche Bank and participated in an international consortium studying large-scale security threats in telecommunication networks. Furthermore, I cooperated with the Computer Security Group of University of California, Santa Barbara in several projects, including a detection system for evasive web-malware.
If you have any fun project ideas, feel free to drop me a line.
Vasilios Mavroudis
v.mavroudis at ucl.ac.ukComputer Science Department
University College London
Gower Street
London WC1E 6BT
United Kingdom
