4 papers
When No Benchmark Exists: Validating Comparative LLM Safety Scoring Without Ground-Truth Labels
Sushant Gautam, Finn Schwall, Annika Willoch Olstad +6
Many deployments must compare candidate language models for safety before a labeled benchmark exists for the relevant language, sector, or regulatory regime. We formalize this sett…
Steering grids for sparse-autoencoder features: when a top-context label names an activation regime rather than a causal axis
Michael A. Riegler, Birk Sebastian Frostelid Torpmann-Hagen
The standard protocol for interpreting sparse-autoencoder (SAE) features labels each feature from its top-activating contexts and validates the label by steering that single featur…
Defending against Stegomalware in Deep Neural Networks with Permutation Symmetry
Birk Torpmann-Hagen, Michael A. Riegler, PÃ¥l Halvorsen +1
Deep neural networks are being utilized in a growing number of applications, both in production systems and for personal use. Network checkpoints are as a consequence often shared…
Probabilistic Runtime Verification, Evaluation and Risk Assessment of Visual Deep Learning Systems
Birk Torpmann-Hagen, PÃ¥l Halvorsen, Michael A. Riegler +1
Despite achieving excellent performance on benchmarks, deep neural networks often underperform in real-world deployment due to sensitivity to minor, often imperceptible shifts in i…