AI
Witeness Overlap: Directional Provenance Inside Open-Weight Model Families
Publication: cs.AI updates on arXiv.org
Author: Siyuan Li, Haoxuan Zeng, Xin Luo, Fernando Jia, Florence Li, Zhengyang Geng, Zico Kolter, Tai Sing Lee, Tianqin Li
Source published: 29/09/2026, 04:00:00 UTC
Briefing created: 29/09/2026, 05:25:34 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 29/09/2026, 05:18:23 UTC.
Briefings
A new white-box test called Witness Overlap uses a third checkpoint as a witness to determine the direction of provenance between two models. The study tested 176 checkpoints across 16 families using Frobenius cosine similarity, achieving 95.3% accuracy in distinguishing parent-child relationships. The method is prompt-free and training-free, offering robustness against weight noise and sparse pruning.
Key points
- The method introduces a local geometric comparison using a third checkpoint as a witness to infer directionality in model provenance.
- Witness Overlap achieves 95.3% accuracy in orienting parent-child relationships using Frobenius cosine similarity across 176 checkpoints from 16 families.
Context and limitations
The study covers only 176 checkpoints from 16 families; results may not generalize to VLMs, diffusion models, or other architectures not included in the evaluation.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 29/09/2026, 05:25:34 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeidxzaoiaws53revlodnmzrsuiykox3nxwednmi7ld645ur3lanove