Writing · Notes
Blog
Research, engineering, and ideas.
June 7, 2026 · 8 min read
QACD: Corrupting Exactly What the Question Asks
Query-Aware Contrastive Decoding turns a vision-language model into a planner that designs a targeted, localized image corruption, attacking the specific evidence a question depends on, to reduce object hallucination without any training.
AIcomputer visionNLPhallucinationresearch
April 28, 2026 · 7 min read
Self-Aug: Teaching Vision Models to Pick Their Own Medicine
A literature survey on Self-Aug, a contrastive decoding method that lets Large Vision-Language Models choose query-specific image corruptions to reduce hallucination.
AIcomputer visionNLPhallucinationresearch
April 1, 2026 · 2 min read
Hello World — Welcome to My Blog
Introducing this space: a place for thoughts on AI, engineering, philosophy, and the things I learn along the way.
intropersonal