🎓 PhD Researcher at Eötvös Loránd University (ELTE) Working on medical and biological AI problems from the Theoretical Physics department. It's a long story, let's say I love AI for Science :D
🔬 Research interests: Deep Learning, AI for Science, Scientific Machine Learning, Generative Models, and Deep Learning for Scientific & Medical Imaging.
🧑💻 Old-school coder, new-school workflow: I can code from scratch, but I also know how to use LLMs to vibe-code effectively.
🧠 My research focuses on applying and developing modern deep learning methods for scientific problems, including:
- Generative and representation learning
- Transformers, GANs, Diffusion Models & Flow Matching
- 3D scientific and medical imaging
- Computational biology and biological representation learning
- Physics-informed and physics-guided machine learning
💻 I work extensively with PyTorch, distributed multi-GPU training, HPC environments, and reproducible ML pipelines.
🏆 Co-Founder and current leader of our faculty deep learning competiton team. link It's not just about competition, it's about building university AI community.
👨🏫 Mentoring I teach computational methods ranging from Python and C/C++ to machine learning and modern generative AI, including Transformers, LLMs, GANs, Diffusion Models and Flow Matching. I also supervise and mentor students working on machine learning and computational research projects.
I have co-designed and taught courses/material. My currently ongoing lectures are:
📚 Selected Research
Check out my repositories below for implementations of my research projects and reproducible experiments.
FLORA — Deep learning for 3D proton imaging and scientific reconstruction. The project explores generative and physics-guided approaches for CT-to-pCT prediction, including latent representation learning and flow-matching models.
single_cell — Generative and representation learning for single-cell perturbation modeling, using Transformer-based representations and latent adversarial learning. The main focus is preserving biological diversity while maintaining predictive accuracy.
nr_dose - Deep learning for proton dose prediction, combining scientific/physical information with neural models for 3D dose estimation.
Some of my current research projects are not publicly available yet:
- Histopathology Vision-Language Models — developing and evaluating multimodal models for computational pathology. The project is currently being developed locally and will be made public when appropriate.
- Agentic AI for Scientific Research — developing human-in-the-loop AI assistants to support research workflows to our faculty, including coding and scientific writing.
Deep Learning PyTorch · Transformers · GANs · Diffusion Models · Flow Matching · Representation Learning · Math
Scientific ML Scientific Computing · Inverse Problems · Computational Imaging · Physics-Informed ML
Infrastructure CUDA · Multi-GPU Training · Distributed Training · HPC · Singularity
Programming Python · C/C++ · Git · Linux
📫 Contact
- Email: bence.dudas20@gmail.com
- Academic: dudas.bence@ttk.elte.hu