Theory for Deep Learning & AI
Building mathematical frameworks for deep networks and AI systems, with an emphasis on theoretical guarantees for their strong performance.
Welcome!
I am a PhD student in the Department of Computer and Information Science at the University of Pennsylvania, advised by René Vidal. I am affiliated with the IDEAS, ASSET and GRASP Center at UPenn. My research interests lie in the theoretical foundations of deep learning, (multi-agent) optimization and theoretical computer science.
Before coming to the United States, I completed a Diploma in Electrical and Computer Engineering in Greece, where I worked with Dimitris Fotakis and Pavlos Efraimidis on problems at the intersection of opinion dynamics and theoretical computer science.
Recent News
June 2026
New preprint on SGD with large stepsizes!
February 2026
Two papers accepted at ICLR 2026!
September 2025
Visiting Simons NYC for Deep Learning Theory. Happy to connect and discuss research!
Honors & Awards
🥉
International Mathematics Competition (IMC)
Bronze Medal
🏅
IEEEXtreme 13.0 Collegiate Programming Competition
Top 10% Worldwide
🥉
South Eastern European Mathematical Olympiad (SEEMOUS)
Bronze Medal (2x)
Research
Optimization Dynamics
Understanding the behavior of stochastic algorithms in modern AI regimes, including stability, implicit bias, and acceleration.
Robust and Certified learning
Designing classifiers and latent-space methods with principled robustness guarantees.
Publications
2026
SGD at the Edge of Stability: Stochastic Stabilization with Large Learning Rates
Shuffling the Data, Stretching the Step-size: Sharper Bias In Constant Step-size SGD
In the 14th International Conference on Learning Representations (ICLR 2026).
ICLR 2026
SSCD: Sparse Semantic Concept Defense Against Semantic Adversarial Attacks to Image Classifiers
In the Unifying Concept Representation Learning Workshop, ICLR 2026.
UCRL Workshop, ICLR 2026
2025
Certified Robustness from Approximate Gaussian
Mixture Structures in Pretrained Latent Spaces
Leveraging knowledge about the underlying distribution to establish robust and certified classifiers.
GreeksInAI Symposium
2024
Earlier
Opinion Dynamics in Graphs with Hidden Links
Recovering the weights of a directed graph on which opinion dynamics are diffused. Formulated the necessary conditions for the uniqueness of a solution and established the number of samples needed to recover the underlying graph.
Undergraduate Thesis
Professional Service
Reviewing
ICML (2024, 2026) · AISTATS (2025, 2026) · ICLR 2026 · NeurIPS 2026 · JMLR
Organizing
Workshop on Machine Learning and Optimization, IEEE CISS 2023
Area Chair
New Frontiers in Game-Theoretic Learning - NExT-Game, ICML 2026
Volunteer
DeepMath 2024
Contact