5 papers
Revealing Hidden Model Behaviors with Task-Specific Self-Reports
Taras Kutsyk, Bartosz ZieliÅski
Fine-tuning can give a language model a hidden behavior--it may give false answers under a narrow condition, or give harmful advice only when a prompt touches a particular topic. W…
Beyond [cls]: Exploring the true potential of Masked Image Modeling representations
Marcin PrzewiÄźlikowski, Randall Balestriero, Wojciech JasiÅski +2
Masked Image Modeling (MIM) has emerged as a promising approach for Self-Supervised Learning (SSL) of visual representations. However, the out-of-the-box performance of MIMs is typ…
Parameter-Efficient Interventions for Enhanced Model Merging
Marcin Osial, Daniel Marczak, Bartosz ZieliÅski
Model merging combines knowledge from task-specific models into a unified multi-task model to avoid joint training on all task data. However, current methods face challenges due to…
OMENN: One Matrix to Explain Neural Networks
Adam Wróbel, MikoÅaj Janusz, Bartosz ZieliÅski +1
Deep Learning (DL) models are often black boxes, making their decision-making processes difficult to interpret. This lack of transparency has driven advancements in eXplainable Art…
Revisiting FunnyBirds evaluation framework for prototypical parts networks
Szymon OpÅatek, Dawid Rymarczyk, Bartosz ZieliÅski
Prototypical parts networks, such as ProtoPNet, became popular due to their potential to produce more genuine explanations than post-hoc methods. However, for a long time, this pot…