Di Fu

VideoResearcher achieves state-of-the-art performance among self-improving agents and approaches the human-designed upper bound, demonstrating a training-free paradigm for long-video understanding that expands agent capabilities through autonomous tool development while reducing costly manual engineering.

Focusing on relevant clue intervals can significantly boost long-video understanding accuracy while reducing the input frame burden on models.

Cooperative multi-agent training can unlock unsupervised reasoning capabilities in RL, yielding performance gains that rival supervised methods without the need for costly annotations.

Subtracting information from the student rather than adding it to the teacher can yield performance improvements that rival those achieved with privileged data.

U-OPSD enables LLMs to self-improve without any external supervision, achieving up to 10.7% performance gains on challenging reasoning benchmarks.

VCSD achieves up to a 4.5% performance boost over traditional methods by distilling knowledge without requiring external teachers or visual cues.