Afroditi Papadaki

Lead Data Scientist and AI Researcher

About Me

I am a researcher in trustworthy machine learning. I study how models behave, how to measure that behaviour and design methods to improve it. I have developed principled approaches to fairness and robustness in federated learning and currently I study the same questions in language models.

I am a Lead Data Scientist at L&G, where I design and build generative and agentic AI systems for financial and enterprise use. Part of that work is evaluation and guardrails, including hallucinations and grounding detection, and factuality checks for systems in production. I am an Area Chair for ICML and NeurIPS and I review for ICLR, AISTATS, KDD, and TMLR.

Before that, I was a Researcher and Lecturer in Data Science and Signal Processing at UCL, where I did my PhD on fair federated learning under Prof. Miguel Rodrigues.

Interests

Trustworthy & Responsible AI
Robustness & Reliability of LLMs
Federated Learning
Evaluation of Generative Models

Education

PhD in Artificial Intelligence
University College London
MSc in Internet Engineering
University College London
BSc Digital Systems
University of Piraeus

Featured Publications

Large Language Models

Generating Diverse Negations from Affirmative Sentences

A method for generating diverse, semantically controlled negations from affirmative sentences, probing and improving how language models handle negation.

Nov 19, 2024

Federated Learning

Federated Fairness without Access to Sensitive Groups

Group fairness in federated learning when sensitive attributes cannot be observed: we protect worst-case subgroups via adversarially learned weightings, with guarantees and experiments on vision and tabular benchmarks.

Feb 22, 2024

Federated Learning

Fair Federated Learning

Approaches for fairness and robustness in decentralized learning — with and without access to sensitive attributes — with formal guarantees, evaluated on computer vision and tabular data.

Oct 1, 2023

Federated Learning

Minimax Demographic Group Fairness in Federated Learning

We formulate group fairness in federated learning as a minimax optimization over demographic group risks and provide algorithms with formal performance guarantees — protecting the worst-off group across heterogeneous clients.

Jun 20, 2022

Robustness

Blind pareto fairness and subgroup robustness

A minimax framework that protects worst-case, unknown subgroups without demographic labels, trading off between Pareto efficiency and blind subgroup robustness.

Jun 10, 2021

Recent Publications

Darian Rodriguez Vasquez, Afroditi Papadaki (2024). Generating Diverse Negations from Affirmative Sentences. In AFM at NeurIPS 2024.

PDF Cite Code Poster

Afroditi Papadaki, Natalia Martinez, Martin Bertran, Guillermo Sapiro, Miguel Rodrigues (2024). Federated Fairness without Access to Sensitive Groups. In International Workshop on Federated Learning: Recent Advances and New Challenges at NeurIPS 2022 (extended version on arXiv, 2024).

PDF Cite

Afroditi Papadaki (2023). Fair Federated Learning. An open access version available from UCL Discovery.

PDF Cite

Afroditi Papadaki, Natalia Martinez, Martin Bertran, Guillermo Sapiro, Miguel Rodrigues (2022). Minimax Demographic Group Fairness in Federated Learning. In FAccT 2022.

PDF Cite Code

Natalia Martinez, Martin Bertran, Afroditi Papadaki, Miguel Rodrigues, Guillermo Sapiro (2021). Blind pareto fairness and subgroup robustness. In ICML 2021.

PDF Cite Code Video