CEDO, a combination of modality-specific learning rates, Pareto gradient synergy, and loss rescaling, reports improved accuracy on five Med-VQA benchmarks including two new biased splits.
General greedy de-bias learn- ing.IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 45:1–17,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Cause-Effect Driven Optimization for Robust Medical Visual Question Answering with Language Biases
CEDO, a combination of modality-specific learning rates, Pareto gradient synergy, and loss rescaling, reports improved accuracy on five Med-VQA benchmarks including two new biased splits.