Publications

For a full list, see my Google Scholar profile.

2026

  1. arXiv
    Certification-Based Differentially Private Learning
    Mihnea Ghitu and Matthew Wicker
    arXiv preprint, 2026 · under review
    arXiv PDF
    @article{ghitu2026certification,
  title   = {Certification-Based Differentially Private Learning},
  author  = {Ghitu, Mihnea and Wicker, Matthew},
  journal = {arXiv preprint arXiv:2609.39629},
  year    = {2026}
}
  2. arXiv
    Towards Poisoning Robustness Certification for Natural Language Generation
    Mihnea Ghitu and Matthew Wicker
    arXiv preprint, 2026 · under review
    arXiv PDF
    @article{ghitu2026poisoning,
  title   = {Towards Poisoning Robustness Certification for Natural Language Generation},
  author  = {Ghitu, Mihnea and Wicker, Matthew},
  journal = {arXiv preprint arXiv:2602.09757},
  year    = {2026}
}

2025

  1. TMLR
    Model Guidance via Robust Feature Attribution
    Mihnea Ghitu, Vihari Piratla, and Matthew Wicker
    Transactions on Machine Learning Research (TMLR), 2025
    OpenReview PDF arXiv
    @article{ghitu2025model,
  title   = {Model Guidance via Robust Feature Attribution},
  author  = {Ghitu, Mihnea and Piratla, Vihari and Wicker, Matthew Robert},
  journal = {Transactions on Machine Learning Research},
  issn    = {2835-8856},
  year    = {2025},
  url     = {https://openreview.net/forum?id=AVAHxDSqUu}
}

2024

  1. Thesis
    Robustness–Privacy Trade-off in Bayesian Neural Networks
    Mihnea Ghitu
    MEng thesis, Imperial College London, 2024 · Distinguished Project
    PDF Code

2023

  1. ICLP
    PySpArX – A Python Library for Generating Sparse Argumentative eXplanations for Neural Networks
    Ioana Mihailescu*, Alan Weng*, Siddharth Sharma*, Mihnea Ghitu*, Dilshaan Grewal*, Khemi Chew*, Hamed Ayoobi, Nico Potyka, and Francesca Toni
    Proceedings of the 39th International Conference on Logic Programming (ICLP 2023), System Demonstrations Track · * equal contribution
    PDF
    @inproceedings{mihailescu2023pysparx,
  title     = {{PySpArX} -- A Python library for generating Sparse Argumentative eXplanations for neural networks},
  author    = {Mihailescu, Ioana and Weng, Alan and Sharma, Siddharth and Ghitu, Mihnea and Grewal, Dilshaan and Chew, Khemi and Ayoobi, Hamed and Potyka, Nico and Toni, Francesca},
  booktitle = {Proceedings of the 39th International Conference on Logic Programming (ICLP 2023)},
  pages     = {336--336},
  publisher = {Open Publishing Association},
  year      = {2023}
}