Code for our work on Distribution Inference
-
Updated
Jul 19, 2024 - Jupyter Notebook
Code for our work on Distribution Inference
Code for our paper 'Formalizing Distribution Inference Risks'
Passive task-provenance attack on merged LoRA adapters: from the published weights alone (no queries, no metadata), spectral features reveal which task families went into a merge (AUC 0.97). Includes two defenses and why fixed-attacker evaluation overstates them. Code and results for the ATTRIB workshop paper, NeurIPS 2026.
To associate your repository with the property-inference topic, visit your repo's landing page and select "manage topics."