Home Team Publications PhD Info

REAL Lab

Our research is dedicated to making AI systems transparent, trustworthy, and accountable.

We develop principled methods to explain why AI models make the decisions they do, and to provide rigorous guarantees when those decisions matter most. Our work spans counterfactual explanations, mechanistic interpretability, privacy in generative models, and robust XAI.

↗ Lab Director ↗ Centre for Explainable AI ↗ ConTrust Project

News

Jan 2026

Paper Paper accepted at ICLR 2026

Our work on synthesising counterfactual explanations via label-conditional Gaussian Mixture VAEs will appear at ICLR 2026.

Nov 2025

Workshop XAISEC workshop accepted at EuroS&P 2026

Our workshop on Explainable AI and Computer Security (XAISEC) was accepted at IEEE EuroS&P 2026.

Sep 2025

Paper Paper on LLM representation consistency at NeurIPS 2025

Our paper on representation consistency for LLMs was accepted at NeurIPS 2025. arXiv →

Sep 2025

Team Dr Leofante joins Imperial as Assistant Professor

Francesco Leofante officially joined the Department of Computing at Imperial College London as Assistant Professor.

Aug 2025

Paper Oral at INTERSPEECH 2025

Mansi presented our work on dementia speech alignment with diffusion-based image generation as an oral at INTERSPEECH 2025.

Apr 2025

Paper Three papers accepted at IJCAI 2025

Three papers from our group will be presented at IJCAI 2025 in Montreal.

Mar 2025

Book Robust Explainable AI published

Our book on Robust Explainable AI is now available from Springer. Springer →

Nov 2024

Paper Paper accepted in AIJ

Our work on probabilistically robust counterfactual explanations under model changes was accepted in the journal of Artificial Intelligence (AIJ).

Featured Research

A selection of ongoing research threads in the lab.

Robust Counterfactual Explanations

We develop counterfactual explanations that remain valid under model changes, data shifts, and adversarial perturbations, providing provable guarantees using formal verification and robust optimisation.

XAI Formal Methods Robustness
See related papers →

Selective Fine-Tuning for Targeted and Robust Concept Unlearning

We localize concept neurons in the cross attention layers of T2I diffusion models and develop fine-tuning strategies that selectively remove specific concepts from generative models while preserving overall model utility, enabling targeted, robust unlearning without full retraining.

Unlearning Diffusion Models Fine-Tuning
See related papers →

Argumentative AI Explanations

We use computational argumentation to build faithful, self-debated explanations for neural network decisions, bridging the gap between formal logic and human-understandable reasoning.

Argumentation LLMs XAI
See related papers →
We are always looking for new PhD students, Postdocs, and Master students to join the team — see PhD Info!

Current Members

Dr Francesco Leofante

Dr Francesco Leofante

Assistant Professor & Lab Director
Co-Director, Centre for Explainable AI

Web Page Google Scholar
Mansi

Mansi

PhD Student
Interpretability & Privacy in Generative AI

Web Page Google Scholar
Emanuele Albini

Emanuele Albini

PhD Student
Machine Learning & Explainable AI

Web Page Google Scholar

Alumni

— (Alumni will be listed here as the lab grows)

Publications

Publications from REAL Lab members. See also Francesco's Google Scholar and Mansi's Google Scholar.

Live data from Google Scholar · Last updated 22 May 2026 · Ordered by citations per year

2026

Selective Fine-Tuning for Targeted and Robust Concept Unlearning
Mansi, A Kori, F Toni, S Demetriou
arXiv preprint arXiv:2602.07919, 2026
View →
Attribution-based Explanations for Markov Decision Processes
P Kobialka, A Pferscher, F Leofante, E Ábrahám, SLT Tarifa, EB Johnsen
Proceedings of IJCAI 2026, 2026
View →
Evaluating Counterfactual Explanation Methods on Incomplete Inputs
F Leofante, D Neider, M Yalçıner
arXiv preprint arXiv:2604.08004, 2026
View →
Probabilistically Robust Counterfactual Explanations under Model Changes
L Marzari, F Leofante, F Cicalese, A Farinelli
Artificial Intelligence, 104459, 2026
View →
Synthesising Counterfactual Explanations via Label-Conditional Gaussian Mixture Variational Autoencoders
J Jiang, F Leofante, A Rago, F Toni
Proceedings of ICLR 2026, 2026
View →

2025

On the Impact of Sparsification on Quantitative Argumentative Explanations in Neural Networks
D Peacock, Mansi, P Nico, T Francesca, Y Xiang
3rd International Workshop on Argumentation for eXplainable AI, 2025
View →
Understanding Dementia Speech Alignment with Diffusion-Based Image Generation
Mansi, A Lepipas, D Woszczyk, Y Guan, S Demetriou
INTERSPEECH, 2025, 2025
View →
Leaky Diffusion: Attribute Leakage in Text-Guided Image Generation
A Lepipas, M Charalambides, J Liu, Y Guan, DC Woszczyk, Mansi, TH Le, ...
Proceedings on Privacy Enhancing Technologies, 2025
View →
Explainable AI, energy and critical infrastructure systems
F Leofante, A Artelt, D Eliades, A Korre, F Toni, T Miller
AI Magazine 46 (3), e70033, 2025
View →
Out-of-Distribution Detection using Counterfactual Distance
M Stoica, F Leofante, A Lomuscio
arXiv preprint arXiv:2508.10148, 2025
View →
Argumentative Ensembling for Robust Recourse under Model Multiplicity
J Jiang, A Rago, F Leofante, F Toni
arXiv preprint arXiv:2506.20260, 2025
View →
Representation Consistency for Accurate and Coherent LLM Answer Aggregation
J Jiang, T Bewley, SI Amoukou, F Leofante, A Rago, S Mishra, F Toni
Proceedings of NeurIPS 2025, 2025
View →
Robust Explainable AI
F Leofante, M Wicker
Springer, 2025
View →
Counterfactual Strategies for Markov Decision Processes
P Kobialka, L Gerlach, F Leofante, E Ábrahám, SLT Tarifa, EB Johnsen
Proceedings of IJCAI 2025, 2025
View →
RobustX: Robust Counterfactual Explanations Made Easy
J Jiang, L Marzari, A Purohit, F Leofante
Proceedings of IJCAI 2025, 2025
View →
Explaining Reinforcement Learning Policies for Power Grid Operations
L Marzari, F Leofante, E Marchesini
View →
Counterfactual Scenarios for Automated Planning
N Gigante, F Leofante, A Micheli
Proceedings of KR 2025, 2025
View →
Counterfactual Explanations Under Model Multiplicity and Their Use in Computational Argumentation
G Alfano, A Gould, F Leofante, A Rago, F Toni
Proceedings of IJCAI 2025, 2025
View →
Explainable AI: definition and attributes of a good explanation for health AI
E Kyrimi, S McLachlan, J Wohlgemut, Z Perkins, D Lagnado, W Marsh, ...
AI Ethics 5 (4), 3883--3896, 2025
View →
Explainability in Machine Learning: Preliminaries and Overview
F Leofante, M Wicker
Robust Explainable AI, 5-15, 2025
View →
Robustness of Counterfactual Explanations
F Leofante, M Wicker
Robust Explainable AI, 17-40, 2025
View →
Robustness of Saliency-Based Explanations
F Leofante, M Wicker
Robust Explainable AI, 41-71, 2025
View →

2024

AmalREC: A Dataset for Relation Extraction and Classification Leveraging Amalgamation of Large Language Models
Mansi, P Pandya, MB Vora, S Bharadwaj, A Anand
arXiv preprint arXiv:2412.20427, 2024
View →
Robust Counterfactual Explanations in Machine Learning: A Survey
J Jiang, F Leofante, A Rago, F Toni
Proceedings of IJCAI 2024, 2024
View →
Rigorous Probabilistic Guarantees for Robust Counterfactual Explanations
L Marzari, F Leofante, F Cicalese, A Farinelli
Proceedings of ECAI 2024, 2024
View →
Contestable AI needs Computational Argumentation
F Leofante, H Ayoobi, A Dejl, G Freedman, D Gorur, J Jiang, ...
Proceedings of KR 2024, 2024
View →
Interval Abstractions for Robust Counterfactual Explanations
J Jiang, F Leofante, A Rago, F Toni
Artificial Intelligence 336, 2024
View →
Recourse under Model Multiplicity via Argumentative Ensembling
J Jiang, A Rago, F Leofante, F Toni
Proceedings of AAMAS 2024, 2024
View →
Promoting Counterfactual Robustness through Diversity
F Leofante, N Potyka
Proceedings of AAAI 2024, 2024
View →

2023

Provably Robust and Plausible Counterfactual Explanations for Neural Networks via Robust Optimisation
J Jiang, J Lan, F Leofante, A Rago, F Toni
Proceedings of ACML 2023, 2023
View →
Robust Explanations for Human-Neural Multi-agent Systems with Formal Verification
F Leofante, A Lomuscio
European Conference on Multi-Agent Systems, 244-262, 2023
View →
Verification of semantic key point detection for aircraft pose estimation
P Kouvaros, F Leofante, B Edwards, C Chung, D Margineantu, ...
Proceedings of the International Conference on Principles of Knowledge …, 2023
View →
Counterfactual Explanations and Model Multiplicity: a Relational Verification View
F Leofante, E Botoeva, V Rajani
Proceedings of the 20th International Conference on Principles of Knowledge …, 2023
View →
Verification-friendly Networks: the Case for Parametric ReLUs
F Leofante, P Henriksen, A Lomuscio
International Joint Conference on Neural Networks (IJCNN'23), 2023
View →
OMTPlan: A Tool for Optimal Planning Modulo Theories
F Leofante
Journal on Satisfiability, Boolean Modeling and Computation 14 (1), 17-23, 2023
View →
Towards robust contrastive explanations for human-neural multi-agent systems
F Leofante, A Lomuscio
Proceedings of the 22nd International Conference on Autonomous Agents and …, 2023
View →
Formalising the Robustness of Counterfactual Explanations for Neural Networks
J Jiang, F Leofante, A Rago, F Toni
Proceedings of AAAI 2023, 2023
View →

2022

Repairing misclassifications in neural networks using limited data
P Henriksen, F Leofante, A Lomuscio
Proceedings of the 37th ACM/SIGAPP Symposium on Applied Computing, 1031-1038, 2022
View →

2021

Formal analysis of neural network-based systems in the aircraft domain
P Kouvaros, T Kyono, F Leofante, A Lomuscio, D Margineantu, ...
International Symposium on Formal Methods, 730-740, 2021
View →

2020

ARCH-COMP20 Category Report: Artificial Intelligence and Neural Network Control Systems (AINNCS) for Continuous and Hybrid Systems Plants
TT Johnson, DM Lopez, P Musau, HD Tran, E Botoeva, F Leofante, ...
EPiC Series in Computing 74, 107-139, 2020
View →
Verification of Neural Networks: Enhancing Scalability through Pruning
D Guidotti, F Leofante, L Pulina, A Tacchella
Proceedings of ECAI'20, 2020
View →
Optimal Planning Modulo Theories
F Leofante, E Giunchiglia, E Abrahám, A Tacchella
Proceedings of IJCAI'20, 4128--4134, 2020
View →

2019

Verification and repair of neural networks: a progress report on convolutional models
D Guidotti, F Leofante, L Pulina, A Tacchella
International Conference of the Italian Association for Artificial …, 2019
View →
Smt-based planning for robots in smart factories
A Bit-Monnot, F Leofante, L Pulina, A Tacchella
International Conference on Industrial, Engineering and Other Applications …, 2019
View →
Repairing learned controllers with convex optimization: a case study
D Guidotti, F Leofante, C Castellini, A Tacchella
International Conference on Integration of Constraint Programming …, 2019
View →
Improving reliability of myocontrol using Formal Verification
D Guidotti, F Leofante, A Tacchella, C Castellini
IEEE Transactions on Neural Systems and Rehabilitation Engineering 27 (4 …, 2019
View →
Integrated synthesis and execution of optimal plans for multi-robot systems in logistics
F Leofante, E Ábrahám, T Niemueller, G Lakemeyer, A Tacchella
Information Systems Frontiers 21 (1), 87-107, 2019
View →
Engineering Controllers for Swarm Robotics via Reachability Analysis in Hybrid Systems
F Leofante, S Schupp, E Abraham, A Tacchella
ECMS International Conference on Modelling and Simulation, 2019
View →

2018

Task planning with OMT: an application to production logistics
F Leofante, E Ábrahám, A Tacchella
International Conference on Integrated Formal Methods, 316-325, 2018
View →
Automated Verification of Neural Networks: Advances, Challenges and Perspectives
F Leofante, N Narodytska, L Pulina, A Tacchella
arXiv preprint arXiv:1805.09938, 2018
View →
Optimal Multi-robot Task Planning: from Synthesis to Execution (and Back)
F Leofante
Proceedings of IJCAI'18, 2018
View →
Guaranteed Plans for Multi-Robot Systems via Optimization Modulo Theories
F Leofante
Proceedings of AAAI'18, 2018
View →

2017

On the Synthesis of Guaranteed-Quality Plans for Robot Fleets in Logistics Scenarios via Optimization Modulo Theories
F Leofante, E Abrahám, T Niemueller, G Lakemeyer, A Tacchella
IEEE International Conference on Information Reuse and Integration 2017 (IRI), 2017
View →
Towards CLIPS-based Task Execution and Monitoring with SMT-based Planning and Optimization
T Niemueller, G Lakemeyer, F Leofante, E Abraham
PlanRob@ICAPS'17, 2017
View →

2016

Learning in physical domains: mating safety requirements and costly sampling
F Leofante, A Tacchella
Conference of the Italian Association for Artificial Intelligence, 539-552, 2016
View →
Combining static and runtime methods to achieve safe standing-up for humanoid robots
F Leofante, S Vuotto, E Ábrahám, A Tacchella, N Jansen
International Symposium on Leveraging Applications of Formal Methods, 496-514, 2016
View →
Autonomous Driving and Undergraduates: an Affordable Setup for Teaching Robotics
N Arnaldi, C Barone, F Fusco, F Leofante, A Tacchella
Italian Workshop on Artificial Intelligence and Robotics, 2016
View →
Learning with safety requirements: state of the art and open questions
F Leofante, L Pulina, A Tacchella
International Workshop on Experimental Evaluation of Algorithms for Solving …, 2016
View →

PhD Information

Pursuing a PhD with REAL Lab

We welcome applications from outstanding students with a strong background in computer science, mathematics, or a related field. PhD students in our lab work on cutting-edge problems at the intersection of explainability, interpretability, and AI safety — with direct impact on how AI systems are understood and deployed in the real world.

Current PhD topics in our group include:

  • Robust and formal counterfactual explanations
  • Argumentation-based XAI and contestable AI
  • Mechanistic interpretability of diffusion models and LLMs
  • Privacy and attribute leakage in generative AI
  • XAI for critical infrastructure systems (energy)

Funded studentships may be available through various EPSRC schemes and industrial partnerships. Please get in touch to discuss potential topics and funding options.

Contact Dr Leofante ↗

How to Apply

Applicants should have (or expect to receive) a first-class or upper second-class degree (or equivalent) in a relevant discipline. A strong mathematical background and programming experience are essential. Experience with machine learning and deep learning frameworks is highly desirable.

To apply, please send an email to Dr Leofante including your CV, academic transcripts, a brief research statement (one page), and the names of two referees. Applicants are encouraged to review our recent publications before reaching out.

See Full Application Details ↗