MMBoundary trains multimodal LLMs to express per-step natural-language confidence and uses reinforcement learning with three rewards to calibrate those statements, cutting calibration error by 7.5% and boosting task accuracy by up to 8.3%.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
MMBoundary: Advancing MLLM Knowledge Boundary Awareness through Reasoning Step Confidence Calibration
MMBoundary trains multimodal LLMs to express per-step natural-language confidence and uses reinforcement learning with three rewards to calibrate those statements, cutting calibration error by 7.5% and boosting task accuracy by up to 8.3%.