Szilvia Ujváry

Szilvia Ujváry

Final-year ML PhD student @ Cambridge

Hi! I’m a final-year PhD student in the ML@CL group at the University of Cambridge, supervised by Ferenc Huszár. I study language model generalisation and inductive biases, with a focus on out-of-distribution compositional generalisation, implicit Bayesian inference, and how training data shapes model behaviour.

Alongside my research, I have hands-on industry experience in LLM pre-training at Apple Machine Learning Research, hosted by Marco Cuturi, and post-training at Reasonable, where I led fine-tuning projects in collaboration with NVIDIA. Previously, I did an MPhil in Machine Learning and Machine Intelligence at Cambridge and a BSc in Mathematics at Imperial College London.

Email / Google Scholar / GitHub / LinkedIn / CV

Publications

* equal contribution

Technical reports

Talks

  • LaCy: What Small Language Models Can and Should Learn is Not Just a Question of Loss. Invited talk, Hungarian Machine Learning Days, 2026
  • LLMs, Implicit Bayesian Inference and Compositional Generalisation. Invited talk, NLIP Seminar Series, University of Cambridge, 2025
  • Position: Understanding LLMs Requires More Than Statistical Generalization. Invited talk, Hungarian Machine Learning Days, 2023

Experience

  • Reasonable, Budapest — Student Researcher, LLM post-training (2026). Led post-training in two NVIDIA Nemotron early-access projects, fine-tuning 30B–120B parameter models to write formally verified Rust proofs.
  • Apple Machine Learning Research, Paris — Research Scientist Intern, hosted by Marco Cuturi (2025–2026). Pre-training data selection to reduce hallucinations in small language models.
  • Lakestar, London — Deep Tech Fellow, part-time (2024)
  • University of Cambridge — Research Assistant with Ferenc Huszár (2022)

Education

  • University of Cambridge — PhD in Machine Learning (2023–2027, expected). Visiting student in the Computational and Biological Learning Lab with José Miguel Hernández-Lobato.
  • University of Cambridge — MPhil in Machine Learning and Machine Intelligence (2022–2023). Distinction, ranked 2nd of 44, best dissertation in cohort.
  • Imperial College London — BSc in Mathematics (2019–2022). First Class Honours, Dean’s List in all three years.

Awards & service

  • Citadel PhD Summit 2026: Third Prize and a £4,000 research grant
  • ICML 2026 Silver Reviewer award; reviewer for ICML, NeurIPS, ICLR
  • Emmanuel College, Cambridge: full MPhil scholarship (2022)
  • G-Research Prize, Imperial College London (2021)
  • Marjorie McDermott Scholarship, Imperial College London (2020–2022)