A new framework for multi-task unlearning has been developed, addressing the challenges of removing specific training data from models that operate across multiple tasks. This method significantly reduces interference between tasks while maintaining model performance, achieving notable improvements in unlearning efficiency on computer vision benchmarks.