PhD student, Institute of automation, Chinese academy of science
1 paper at NeurIPS 2025
We accelerates attention flow computation by pruning attention graphs, achieving high efficiency while preserving interpretability.
