Swift Sampling: Selecting Temporal Surprises via Taylor Series

Open paper

Benchmark	Model	Rank	Results
video-question-answering-on-mlvu	Swift Sampling (LLaVA-Video)	–	Accuracy: 67.2