Welcome!

I am Efstathia Soufleri, a Postdoctoral Researcher at the Archimedes Unit of the Athena Research Center.
Before that, I earned my Ph.D. in Electrical and Computer Engineering at Purdue University’s Neuro(nano) Research Laboratory, advised by Prof. Kaushik Roy.

About me

I work on making machine learning models more trustworthy — understanding what they memorize, whether that leaks private data, and how to evaluate them rigorously before deployment. My research spans theoretical work on memorization and privacy attacks, differentially private generative modeling, and evaluation pipelines for large language models in low-resource domains.

Research interests: Trustworthy ML · Differential Privacy · Memorization · LLM Evaluation · Model Auditing

I hold a B.Sc. in Mathematics (Applied Mathematics specialization, National and Kapodistrian University of Athens, 2016) and an M.Sc. in Computer Science (University of Thessaly, 2017). I then pursued my Ph.D. at Purdue University, focusing on efficient and privacy-preserving machine learning algorithms under the supervision of Prof. Kaushik Roy.

Selected work

• Curvature Clues: Decoding Deep Learning Privacy with Input Loss Curvature — NeurIPS 2024, Spotlight
• Memorization Through the Lens of Sample Gradients — ICLR 2026
• MultiFinBen: Benchmarking LLMs for Multilingual and Multimodal Financial Application — ACL 2026
• DP-CARE: A Differentially Private Classifier for Mental Health Analysis — Frontiers in Digital Health, 2025 (mentee-led)