Cite
The dissertation and every paper behind it, in four formats. Individual results and charts are citable too: each carries a permanent link back to the exact view, its population, and its source file.
Sadanandan, Binesh (2026). Paraphrase Sensitivity in Medical Vision-Language Models: Measurement, Mechanisms, Mitigation, and Deployment Safety [Doctoral dissertation, University of New Haven] [Pre-defense draft, 2026-07-15]. https://bineshkumar.me/phd-thesis/
Binesh Sadanandan, "Paraphrase Sensitivity in Medical Vision-Language Models: Measurement, Mechanisms, Mitigation, and Deployment Safety," Ph.D. dissertation pre-defense draft, University of New Haven, West Haven, Connecticut, 2026. [Online]. Available: https://bineshkumar.me/phd-thesis/
@phdthesis{sadanandan2026paraphrase,
author = {Binesh Sadanandan},
title = {Paraphrase Sensitivity in Medical Vision-Language Models: {M}easurement, {M}echanisms, {M}itigation, and {D}eployment {S}afety},
school = {University of New Haven},
year = {2026},
address = {West Haven, Connecticut},
type = {Doctoral dissertation},
note = {Pre-defense draft, 2026-07-15},
url = {https://bineshkumar.me/phd-thesis/}
}
Binesh Sadanandan. Paraphrase Sensitivity in Medical Vision-Language Models: Measurement, Mechanisms, Mitigation, and Deployment Safety. Doctoral dissertation (Pre-defense draft, 2026-07-15), University of New Haven, West Haven, Connecticut, 2026. https://bineshkumar.me/phd-thesis/
Work under review is cited as an unpublished manuscript, without a venue that would identify an anonymous submission.
Sadanandan, B., Behzadan, V., Jayan, L., & Kurup, A. G. (2026). PSF-Med: A clinician-audited benchmark for paraphrase sensitivity in medical vision-language models. MMFM-BIOMED Workshop, CVPR 2026. arXiv:2602.21428
Sadanandan, B., & Behzadan, V. (2026). Mechanistically guided LoRA improves paraphrase consistency in medical vision-language models. Conference on Health, Inference, and Learning (CHIL) 2026. arXiv:2603.00148
Sadanandan, B., & Behzadan, V. (2026). Consistent but dangerous: Per-sample safety classification reveals false reliability in medical vision-language models. MedReasoner Workshop, CVPR 2026. arXiv:2603.20985
Sadanandan, B., & Behzadan, V. (2026). Predictive entropy as a joint screen for error and paraphrase instability in medical vision-language models. Manuscript under review.
Sadanandan, B., & Behzadan, V. (2026). Attention without grounding: Causal evaluation of visual explanations in medical VLMs. Manuscript under review.
Sadanandan, B., & Behzadan, V. (2026). Consistency is not safety: Deployment audits reveal text-driven selection in medical vision-language models. Manuscript under review.
Sadanandan, B., Karimi, A., Upadhayay, B., & Behzadan, V. (2026). Trustworthiness evaluation of medical vision-language models: A scoping review of robustness, grounding, hallucination, and uncertainty. Manuscript under review.
Sadanandan, B., & Behzadan, V. (2026). When chain-of-thought backfires: Evaluating prompt sensitivity in medical language models. 2AI Conference 2026. arXiv:2603.25960
Sadanandan, B., & Behzadan, V. (2025). VSF-Med: A vulnerability scoring framework for medical vision-language models. IEEE ISBI 2026 (pilot abstract). arXiv:2507.00052
Every result and chart on this site has a Cite button. It gives you the thesis chapter, the table or figure, the metric definition, the sample and denominator, the source data file, and the caveat that bounds the number, plus a link that reopens that exact view.
This matters more than usual here. Several numbers in this work are only meaningful with their population attached: a flip rate from the curated diagnostic set is roughly five times the benchmark rate for the same model, and the two are not comparable. A citation that carries the denominator cannot be quietly misread.