Edu-MMBias reveals that vision-language models show compensatory class bias favoring lower-status stories alongside racial and health stereotypes, with visual inputs acting as a backdoor that revives biases despite text-based safeguards.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Edu-MMBias: A Three-Tier Multimodal Benchmark for Auditing Social Bias in Vision-Language Models under Educational Contexts
Edu-MMBias reveals that vision-language models show compensatory class bias favoring lower-status stories alongside racial and health stereotypes, with visual inputs acting as a backdoor that revives biases despite text-based safeguards.