AIBullisharXiv – CS AI · 14h ago7/10
🧠
Mitigating Hallucination in Vision-Language Models through Barrier-Regulated Adaptive Closed-form Steering
Researchers propose BRACS, a training-free framework that reduces hallucinations in vision-language models by monitoring visual grounding during text generation and applying adaptive corrections only when needed. The method achieves significant improvements on hallucination benchmarks while maintaining computational efficiency comparable to baseline decoding speeds.