” It then retrieves a small set of questions with the same skeleton from a lightweight example pool.
Reusable prompt structures help vision-language models interpret negated questions
Skeleton-and-Strategy Prompting identifies a question’s negation pattern, generates a concise answering rule, and reuses that guidance without updating model parameters.
Big Tech
Yuliang Cai · Mohammad Rostami · Jesse Thomason
University of Southern California · Amazon Inc.
Research Digest··3 min read
Cai, Rostami and Thomason introduce Skeleton-and-Strategy Prompting (SSP), an inference-time method for visual question answering involving absence, exclusion and logical opposition.
Why this paper
From Amazon Inc. and University of Southern California
In one line
Skeleton-and-Strategy Prompting improves vision-language models' negation understanding without training by injecting structure-aware prompts.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§