Marvo is a two-agent system.
Self-evolving visual agents learn tool use from unlabeled images using margin reward.
The Marvo framework pairs a question-generating challenger with a reinforcement-learning solver, using the solver's performance gap with and without tools to create an adaptive training curriculum.
Big Tech
Wenxi Yang · Yuzhong Zhao · Baochang Zhang · Yunfan Liu · Qixiang Ye
UCAS · Microsoft Research Asia · Beihang University
Research Digest··2 min read
The authors propose Marvo, a two-agent framework that learns tool invocation for visual agents from unlabeled images.
Why this paper
From Microsoft Research Asia and 2 others
In one line
A margin reward based on the solver's performance difference with and without tools enables visual agents to learn tool invocation from unlabeled images.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§