About
I develop methods that detect when a deep neural network is operating outside its training distribution, and explain why, so that models can be deployed more safely. My work spans out-of-distribution (OOD) detection, runtime monitoring and explainability. I am a PhD candidate at Imperial College London's Safe AI Lab (SAIL), supervised by Prof. Alessio Lomuscio and funded by the UKRI CDT in Safe and Trusted AI.
Before my PhD, I studied computer science at Harvard, spent four years in quantitative roles in commodities and FX markets, and completed an MSc in Advanced Computer Science at Oxford, working on fairness in financial machine learning. During my PhD I have interned at Meta, building LLM tools for data infrastructure, and at Citadel, in quantitative model risk research.
Outside of research, I spent a year playing for the Imperial College Lawn Tennis Club, and I coxed for the Radcliffe Lightweight Women at Harvard and for St Edmund Hall at Oxford. I'm also a PADI-certified scuba diver and have dived in Egypt, Costa Rica and the Caribbean, as well as Iceland's Silfra fissure, where you can swim between two continental plates.
News
- Aug 2026 — Presented Out-of-Distribution Detection using Counterfactual Distance at the IJCAI 2026 Workshop on Explainable AI in Bremen, Germany.
- Jun 2026 — Presented A Robust Out-of-Distribution Detection Framework via Synergistic Smoothing (CVPR Findings) at CVPR 2026 in Denver, Colorado.
- May 2026 — Started a Quantitative Research internship in Model Risk Research at Citadel.
- Jun 2025 — Joined Meta as a PhD intern in Systems and Infrastructure Software Engineering, building LLM tools for data infrastructure.
Publications
Out-of-Distribution Detection using Counterfactual Distance
IJCAI Workshop on Explainable AI 2026
IJCAI 2026 Workshop on Explainable Artificial Intelligence (XAI), Bremen, Germany
BibTeX
@misc{stoica2025counterfactual,
title = {Out-of-Distribution Detection using Counterfactual Distance},
author = {Stoica, Maria and Leofante, F. and Lomuscio, Alessio},
year = {2025},
eprint = {2508.10148},
archivePrefix = {arXiv},
note = {Presented at the IJCAI 2026 Workshop on Explainable Artificial Intelligence (XAI), Bremen, Germany}
}
A Robust Out-of-Distribution Detection Framework via Synergistic Smoothing
CVPR Findings 2026
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Findings
BibTeX
@inproceedings{stoica2026robust,
title = {A Robust Out-of-Distribution Detection Framework via Synergistic Smoothing},
author = {Stoica, Maria and Hekal, A. and Lomuscio, Alessio},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Findings},
year = {2026}
}
Wanted: Personalised Bias Warnings for Gender Bias in Language Models
GeBNLP Workshop 2025
Proceedings of the 6th Workshop on Gender Bias in Natural Language Processing (GeBNLP)
BibTeX
@inproceedings{dibonaventura2025wanted,
title = {Wanted: Personalised Bias Warnings for Gender Bias in Language Models},
author = {Di Bonaventura, C. and Nwachukwu, M. and Stoica, Maria},
booktitle = {Proceedings of the 6th Workshop on Gender Bias in Natural Language Processing (GeBNLP)},
year = {2025}
}
CV
Education
PhD Candidate in Computing
Oct 2022 - Jun 2027 (expected)
Imperial College London
- Research focus: OOD detection and monitoring of neural networks
- Supervisor: Prof. Alessio Lomuscio, Safe Artificial Intelligence Lab
- UKRI CDT in Safe and Trusted AI
MSc in Advanced Computer Science
Oct 2021 - Sep 2022
University of Oxford
- Masters Dissertation: Exploration of Fairness Metrics in Financial Machine Learning
- Supervisor: Prof. Marta Kwiatkowska
- Google DeepMind Scholar
BA in Computer Science
Sep 2013 - May 2017
Harvard College, MIT
Professional Experience
Quantitative Research Intern, Model Risk Research
May 2026 - Nov 2026
Citadel
- Automating validation and monitoring checks for trading and pricing models, and developing independent benchmarks to test them.
PhD Intern, Systems and Infrastructure Software Engineering
Jun 2025 - Sep 2025
Meta
- Developed LLM Model Context Protocol tools to integrate internal data infrastructure knowledge into large language models.
- Built systems enabling LLMs to assist engineers with debugging, querying, and reducing workload on data infrastructure teams.
Foreign Exchange Options Quantitative Developer, Associate
Sep 2019 - Jul 2021
NatWest Markets
- Designed and developed pricing libraries and risk systems for FX Options, working with front and middle office to translate business strategy into technical solutions.
Commodities Sales Strats, Analyst
Jul 2017 - Aug 2019
Goldman Sachs
- Worked with sales and trading to structure and price transactions for clients, and built tools to automate analytical workflows.
Summer Analyst, Strats
Jun 2016 - Aug 2016
Goldman Sachs
Teaching Assistant: Discrete Mathematics
Spring 2017
Harvard College, Department of Computer Science
Leadership
Diversity and Inclusion Committee Member
Jan 2023 - Present
UKRI CDT in Safe and Trusted AI
- Member of the STAI CDT Diversity and Inclusion committee tasked with organizing events to promote diversity and inclusion in the CDT.
Outreach Officer
Sep 2023 - Sep 2025
Women and Non-binary Individuals in Computing, Imperial College London
- Member of the WNBiC committee tasked with organizing outreach events such as school visits and hackathons for girls.
Social Events Co-Chair
Oct 2021 - Sep 2022
University of Oxford Women in Computer Science
- Organized events to promote networking with students and faculty.