When to Rethink: Learning Multi-Perspective Self-Verification for Vision-Language Models
MOTIVE decides whether a vision-language model should trust its first answer or rethink it.
The paper says current self-verification is brittle because results shift with verifier strength and prompt choice. Its proposed framework checks candidate answers from multiple verification perspectives, then learns a reliability score aligned with correctness. At inference time, that score either accepts the answer or triggers history-guided rethinking. The authors report gains over self-verification and self-correction baselines across multimodal benchmarks, with fewer unnecessary reasoning turns. HF Daily Papers' note
The paper says current self-verification is brittle because results shift with verifier strength and prompt choice. Its proposed framework checks candidate answers from multiple verification perspectives, then learns a reliability score aligned with correctness. At inference time, that score either accepts the answer or triggers history-guided rethinking. The authors report gains over self-verification and self-correction baselines across multimodal benchmarks, with fewer unnecessary reasoning turns. HF Daily Papers' note
score 4