Search papers, labs, and topics across Lattice.
Nanjing University
1
0
2
Effective visual tool use hinges on causally grounded supervision, not just imitating tool calls, reshaping how we train multimodal agents.
