> |
Hello 🇵🇱! I'm Łukasz, an ML Researcher at Warsaw University of Technology and IDEAS, working closely with Kamil Deja.
I love diffusion models ❤️. My goal is to better understand the internal representations within generative models and use them to achieve more-precise control over the generation process.
Latest News
Two papers accepted to ICLR 2026: There and Back Again (main track) and TADA (Re-Align Workshop). See you in Rio 🇧🇷!
🎓 Started my PhD at Warsaw University of Technology, co-advised by Tomasz Trzciński and Joost van de Weijer within the ELLIS PhD program.
Joined IDEAS Research Institute as an AI Researcher in the Generative AI Team.
Two papers on Diffusion Models accepted to ICLR 2025: Precise Parameter Localization for Textual Generation (main track) and Low-Rank Continual Adaptation (SCOPE Workshop).
Recent Work
TADA! Tuning Audio Diffusion Models through Activation Steering
We show that various musical concepts are controlled by subset of layers in audio diffusion models and steer activations within those layers for controllable audio generation.
Read more →There and Back Again: On the relation between Noise and Image Inversions in Diffusion Models
Noise approximations produced with inversion methods exhibit input image patterns. We find the root cause and provide simple fix, improving inversion-based image editing, interpolation and style transfer.
Read more →