Metraipolis › archive › dc_714cc7c144fa
FLIP: Final Layer Inference-Time Probing for Vision-Language Models — arXiv:2609.30993
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.30993
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-08T01:39:38.652Z
- SHA-256 of the archived text
815d501a3da51a66afd7555acc7b2787d876586d33cfa601a2955353ea420842- Stable URL of this page
https://metraipolis.com/documento/dc_714cc7c144fa
How to cite
FLIP: Final Layer Inference-Time Probing for Vision-Language Models — arXiv:2609.30993 — arXiv (en), https://arxiv.org/abs/2609.30993 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_714cc7c144fa (Metraipolis archive dc_714cc7c144fa, sha256 815d501a3da51a66afd7555acc7b2787d876586d33cfa601a2955353ea420842)
BibTeX
@misc{dc714cc7c144fa,
title = {FLIP: Final Layer Inference-Time Probing for Vision-Language Models — arXiv:2609.30993},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.30993},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_714cc7c144fa; sha256 815d501a3da51a66afd7555acc7b2787d876586d33cfa601a2955353ea420842},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.30993. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
We present FLIP, a final-layer inference-time probe for testing whether a logit-facing intervention site in an open-weight vision-language model (VLM) supports structured, task-linked computation rather than generic perturbation. Behavioral change under internal intervention is otherwise mechanistically ambiguous: it may reflect improved use of visual evidence, generic output instability, or outright degradation. FLIP applies elementwise flooring to the final normalized hidden state before logit computation, leaving parameters, prompts, and decoding unchanged. On a controlled detection/counting probe, sweeping intervention strength reveals three regions: negligible change, a bounded interior regime in which detection recall at IoU 0.50 ($R_{50}$) improves while tolerant counting error ($\mathcal{E}_{\mathrm{count}}$) falls, and over-suppression. We formalize a four-criterion probe-and-sweep protocol for disciplining the interpretation of intervention effects: regime structure, grounding-proxy alignment, feature-coherence dependence, and failure to reproduce the same positive regime on a performance-based negative control. The post-normalization state passed to the output head is the logit-facing instantiation of this test; under a non-targeted flooring sweep it satisfies the full protocol. Raw decoder-layer interventions, including the last-block output before final normalization, and the singleton-pair left/right control fail to reproduce the Final-site signature, while same-site operators and multiple VLMs replicate it. FLIP is therefore a validation step for intervention-based mechanistic interpretability, not a steering method. Repo at this https URL
Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-28 in cs.AI).
Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.30993 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-08T01:12:14.517Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-28?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).