PhD Candidate in XAI for Time Series at UCD
thesis submitted; viva scheduled
PhD candidate in Computer Science at University College Dublin (UCD), thesis submitted and viva scheduled. My supervisor is Dr. Georgiana Ifrim; we conducted research on Explainable Artificial Intelligence (XAI) for Multivariate Time Series Classification.
To enrich my research and foster international collaboration, I undertook a research visit to the Intelligent System Laboratory at the University of Bristol (England,UK) from April to July 2024, under the supervision of Dr. Zahraa Abdallah.
I'm currently looking for a job, exploring challenging opportunities where I can apply my research skills to complex problems (e.g. Applied Scientist, ML Engineer).
Featured Project: tsCaptum
XAI Library for Time Series Classification
Live on PyPI
Problem: Standard XAI tools (like Captum) are not optimized for time-series data and are computationally slow.
Solution: Built tsCaptum, a Python library that bridges Captum with time-series frameworks (aeon, sktime), reducing computation time by 1-2 orders of magnitude.
Tech stack: Python, PyTorch, Captum, aeon, pytest.
Featured Project: drXAI framework
XAI Library for Time Series Classification
Problem: State of the art for Multivariate Time Series Classifications doesn't scale well with increasing number of channels and time points.
Solution: Built drXAI, a Python framework that selects features (either channels or timepoints) into Multivariate Time Series leveraging on saliency maps. It reduces 80–90% of the data while preserving accuracy.
Tech stack: Python, PyTorch, Captum, tsCaptum.
Publications
Scaling Time Series Classification via XAI-Driven Data Reduction
2026AALTD at PKDD‑ECML 2026
Developed drXAI a framework selecting features into a Multivariate Time Series (either channels or time points) by 1) trainig Hydra, a fast yet effective model 2) apply attribution methods to get feature relevance 3) train SOTA classifiers on reduced data.
An empirical evaluation of factors affecting SHAP explanation of time series classification
2025TempXAI at PKDD‑ECML 2025 / Springer
Found that equal-length segmentation performs best, length-weighted normalization consistently improves quality, and average background yields a small performance boost for SHAP explanations.
View PaperA short tutorial for multivariate time series explanation using tsCaptum
2024Software Impacts, Elsevier
tsCaptum is a library bridging time series frameworks and Captum to efficiently generate saliency maps for univariate/multivariate classifiers using chunking to reduce computational cost.
View PaperImproving the evaluation and actionability of explanation methods for multivariate time series classification
2024PKDD‑ECML 2024 Research Track / Springer
Improved InterpretTime evaluation methodology to work with non-deep learning classifiers, avoid custom training, and be more robust. Experiments confirm SHAP as the best explainer.
View PaperEvaluating explanation methods for multivariate time series classification
2023AALTD at PKDD‑ECML 2023 / Springer
Evaluating explanation methods for multivariate time series classifiers like ROCKET and dResNet using SHAP and dCAM. Results show SHAP outperforms dCAM; techniques for evaluating explainers need refinement.
View PaperEducation
PhD in XAI for Time Series
September 2022 – August 2026University College Dublin, Ireland
Research focus: Explainable Artificial Intelligence (XAI) for Multivariate Time Series Classification
Supervisor: Dr. Georgiana Ifrim
Funding: Science Foundation Ireland (SFI) via ML-Labs research centre
Master of Science in Computer Science
September 2017 – March 2020University of Pisa, Italy
Specialization: Master degree in AI (Machine Learning & Deep Learning)
Final grade: 110/110
Thesis: Image captioning with structure generation
Bachelor of Science in Computer Science
September 2013 – September 2017University of Pisa, Italy
Final grade: 100/110
Work Experience
Research Fellow (Assegno di ricerca)
May 2020 – May 2022University of Pisa, Italy
Applied ML techniques to extract clips from documentary films for a web remix platform. Developed video processing pipelines using OpenCV and FFmpeg.
Teaching Experience
Teaching Assistant
September 2022 – August 2026University College Dublin
• Data Analytics (COMP47350): 2022/23, 2023/24, 2024/25
• Introduction to Artificial Intelligence (COMP30030): 2022/23, 2023/24
• Introduction to Programming I (COMP10010): 2023/24, 2024/25
• Computer Programming II (COMP10120): 2022/23
Skills
- Programming languages: Python, Node.js, MATLAB, Java, JavaScript, C++, C, Bash
- Deep Learning frameworks: PyTorch, TensorFlow with CUDA acceleration
- Time Series: sktime, aeon; SOTA classifiers (ROCKET, MiniRocket, Hydra, WEASEL, InceptionTime, ConvTran)
- XAI: SHAP, KernelSHAP, LIME, Feature Ablation, dCAM; frameworks (Captum, SHAP)
- Other tools: Git, Docker, MongoDB, OpenCV, FFmpeg, Express.js
Job Market
- Role Types: PApplied Scientist, ML Engineer.
- Problem Areas: Time Series problems, XAI, Model Interpretability.
- Time Location: Willing to relocate
Get In Touch
Feel free to reach out for research collaborations, speaking opportunities, or industry positions.