7 papers · 1 filter
Interpretability Without Tradeoffs: Disentangling Polysemanticity At Equal Predictive Performance
DoÄukan BaÄcı, Bernt Schiele, Simone Schaub-Meyer +2
Deep neural networks (DNNs) are widely used, but interpreting what they actually learn remains difficult. A major obstacle is that individual neurons often encode multiple unrelate…
FaCT: Faithful Concept Traces for Explaining Neural Network Decisions
Amin Parchami-Araghi, Sukrut Rao, Jonas Fischer +1
Deep networks have shown remarkable performance across a wide range of tasks, yet getting a global concept-level understanding of how they function remains a key challenge. Many po…
A Granular Study of Safety Pretraining under Model Abliteration
Shashank Agnihotri, Jonas Jakubassa, Priyam Dey +4
Open-weight LLMs can be modified at inference time with simple activation edits, which raises a practical question for safety: do common safety interventions like refusal training…
Spatial Reasoners for Continuous Variables in Any Domain
Bart Pogodzinski, Christopher Wewer, Bernt Schiele +1
We present Spatial Reasoners, a software framework to perform spatial reasoning over continuous variables with generative denoising models. Denoising generative models have become…
Pixel-level Certified Explanations via Randomized Smoothing
Alaa Anani, Tobias Lorenz, Mario Fritz +1
Post-hoc attribution methods aim to explain deep learning predictions by highlighting influential input pixels. However, these explanations are highly non-robust: small, impercepti…
TokenFormer: Rethinking Transformer Scaling with Tokenized Model Parameters
Haiyang Wang, Yue Fan, Muhammad Ferjad Naeem +5
Transformers have become the predominant architecture in foundation models due to their excellent performance across various domains. However, the substantial cost of scaling these…