Resume

I am a machine learning researcher with a PhD in Data Science from Sapienza University of Rome (2026) and more than four years of industry experience as a data scientist at General Electric. From November 2026 I am a postdoctoral researcher at the Free University of Bozen-Bolzano.

My research is on the robustness and evaluation of deep models: learning with noisy labels, graph neural networks, vision-language models, corrective machine unlearning and out-of-distribution generalisation. I design diagnostics that read what a model has learned from its internal representations, and training or post-training interventions that act on what the diagnostics show. My papers appear at EMNLP, ECML PKDD, LoG, IJCNN and ACM SAC. I organised an international competition at IJCNN 2025 and review for ICLR, CVPR and IJCNN.

Education

PhD in Data Science, Sapienza University of Rome

2023 – 2026 · Department of Computer, Control and Management Engineering (DIAG). I defended my thesis, Robust Representation Learning under Imperfect Supervision, on 25 September 2026 (final evaluation: ottimo). My advisor was Prof. Fabrizio Silvestri, and I was funded by a PNRR research scholarship.

MSc in Artificial Intelligence and Robotics, Sapienza University of Rome

2020 – 2023 · CGPA 9.5/10, taught in English. I took courses in deep learning, machine learning, reinforcement learning, natural language processing, vision and perception, planning and reasoning, neuroengineering, human-computer interaction and autonomous mobile robots.

BTech in Information Technology, National Institute of Technology Srinagar

2012 – 2016 · Srinagar, India. CGPA 8.0/10.

Experience

Postdoctoral researcher, Free University of Bozen-Bolzano

From November 2026 · Bolzano, Italy. I join the LLADIGA project (Learning Language with Dialogue Games) with Prof. Raffaella Bernardi. My work there is the robust evaluation of the language-use capabilities of large language models, the analysis of what learning in interaction changes inside the models, and the organisation of community challenges.

Doctoral researcher, Sapienza University of Rome

2023 – 2026 · DIAG, RSTLess group, Rome, Italy. I worked on robustness and evaluation across model families: noise-robust losses, Dirichlet-energy diagnostics for graph neural networks, representation-level evaluation of vision-language models and corrective unlearning. I organised the IJCNN 2025 competition on learning with noisy graph labels, was teaching assistant for the graduate Deep Learning course (2024 – 2025) and co-supervised Master's theses.

Visiting researcher, University of Edinburgh

November 2025 – February 2026 · School of Informatics, Edinburgh, United Kingdom. I was hosted by Dr Pasquale Minervini and worked on the robustness and evaluation of vision-language models under distribution shift, with representation- and frequency-aware diagnostics.

Research-infrastructure collaborator and HPC cluster manager, Sapienza University of Rome

June 2024 – December 2025 · DIAG, Rome, Italy. I managed the department GPU cluster and the SoBigData.it research infrastructure for more than 50 researchers: SLURM scheduling, GPU allocation and reproducible software environments. I won both contracts by comparative evaluation.

Data scientist, General Electric

June 2016 – October 2020 · Bangalore, India. Projects I worked on:

• Defect detection: super-resolution GANs and other deep learning methods to identify defects in turbine shrouds.

• Pricing analytics: gradient-boosted trees to predict customer payment behaviour.

• My Fleet: an analytical service presenting the relative performance of turbines, built with the Spring Framework.

• Wilmington: time-series models to predict peak power consumption at the Wilmington site.

• HR analytics: Transformer-based NLP to validate resumes for specific roles, served through a Flask REST API.

• Customer-query routing with Apriori algorithms, and Tableau dashboards for service agreements.

Software engineer intern, HCL Technologies

November 2014 – February 2015 · Delhi, India. I worked on a demand-management tool built on a microservice architecture (Java, Spring Boot, Maven, REST APIs).

Publications

A current list with citations is on Google Scholar.

Conference and journal papers

• F. A. Wani, A. Suglia, R. Saxena, A. P. Gema, W.-C. Kwan, F. Barez, M. S. Bucarelli, F. Silvestri, P. Minervini. Same Answer, Different Representations: Hidden Instability in VLMs. EMNLP 2026, Main Conference. arXiv:2602.06652

• M. Mozafari, F. A. Wani, M. S. Bucarelli, F. Silvestri. Subtract the Corruption: Training-Data-Free Corrective Machine Unlearning using Task Arithmetic. ECML PKDD 2026, Research Track. arXiv:2511.18660

• A. Ghasemi, F. A. Wani, M. S. Bucarelli, F. Silvestri. PISA: Prioritized Invariant Subgraph Aggregation. Learning on Graphs Conference (LoG) 2026. arXiv:2511.22435

• F. A. Wani, M. S. Bucarelli, G. Di Teodoro, A. G. Di Francesco. IJCNN 2025 Competition: Learning with Noisy Graph Labels. International Joint Conference on Neural Networks (IJCNN) 2025. DOI

• T. M. Wani, M. Uecker, F. A. Wani, I. Amerini. HCN-TA: Hierarchical Capsule Network with Temporal Attention for a Generalizable Approach to Audio Deepfake Detection. ACM/SIGAPP Symposium on Applied Computing (SAC) 2025. DOI

• F. A. Wani, M. S. Bucarelli, F. Silvestri. Learning with Noisy Labels through Learnable Weighting and Centroid Similarity. International Joint Conference on Neural Networks (IJCNN) 2024. DOI · Code

• T. M. Wani, S. A. A. Qadri, F. A. Wani, I. Amerini. Navigating the Soundscape of Deception: A Comprehensive Survey on Audio Deepfake Generation, Detection, and Future Horizons. Foundations and Trends in Privacy and Security, 6(3–4), 153–345, 2024. DOI

Under review

• F. A. Wani, M. S. Bucarelli, A. G. Di Francesco, O. Pryymak, F. Silvestri. The Energy of Memorization: Why Graph-Level Label Noise Sharpens Representations and How to Stop It. arXiv:2412.08419

• F. A. Wani, M. S. Bucarelli, M. H. Mirza, O. Pryymak, A. P. Gema, I. Masi, P. Minervini, F. Silvestri. Cross-Modal Attention Acts as a Frequency Filter: Why Verbose Prompts Improve Robustness in Vision-Language Models.

• F. A. Wani, A. Purificato, M. S. Bucarelli, A. G. Di Francesco, M. Corelli, O. Pryymak, F. Silvestri. Survey and Benchmark of Graph Learning under Label Noise. HAL hal-05698305

PhD thesis

F. A. Wani. Robust Representation Learning under Imperfect Supervision. PhD in Data Science, Sapienza University of Rome, 2026.