Partition-Aware Unlearning for Removing Spurious Correlations in Large Vision-Language Models
The episode discusses a paper titled "Partition-Aware Unlearning for Removing Spurious Correlations in Large Vision-Language Models." The hosts explain how the framework uses structured data partitioning to separate object-grounded examples from background correlations. They detail three partitioning methods and three unlearning objectives used to suppress background reliance while preserving object reasoning, concluding that this approach reduces hallucinations and improves model dependability.