I am a PhD student at Heinrich-Heine-Universität Düsseldorf under the supervision of
Prof. Dr. Paul Swoboda,
working at the intersection of deep learning and natural language processing.
My main research interests include, but are not limited to:
⚡
Improving LLMs without retraining
e.g. prompt engineering and efficient inference-time methods
🎯
Reinforcement learning for LLMs
training models to follow instructions and improve reasoning via RL
🧹
Machine unlearning
selectively removing knowledge from a model after training
Publications
🔥 top-tier AI/ML venue
TATRA: Training-Free Instance-Adaptive Prompting Through Rephrasing and Aggregation
B. Dziuba, K. Kuchta, P. Batorski, P. Spurek, P. Swoboda
Heinrich-Heine-Universität Düsseldorf & Jagiellonian University
Supervisor: Prof. Paul Swoboda
Oct 2020 – Jul 2022
MSc in Mathematics
Jagiellonian University · Graduated with the highest grade
Thesis: Variational autoencoders and their evaluation
Oct 2021 – Feb 2022
Exchange Student
KU Leuven
Oct 2017 – Jul 2020
BSc in Mathematics
Jagiellonian University
Experience
Oct 2022 – Dec 2023
Junior Data Scientist · QuantUp
Contributed to three computer vision projects, one LLM project, and one tabular ML project. Responsible for data preprocessing, modeling, researching optimal solutions, and client-facing presentations.
PythonGitHubAWS
Jul 2022 – Sep 2022
Data Scientist Intern · NorthGravity
Built reusable ML pipelines for time-series forecasting and delivered end-to-end forecasting solutions.
PythonBitbucket
Jul 2020 – Jun 2021
Quantitative Risk Intern · UBS
Developed a comprehensive R package from scratch for time-series forecasting (utilities for model developers, documentation, unit tests). Presented findings to the team via technical talks.
