Track new results across pure and applied mathematics.
Search papers, labs, and topics across Lattice.
5
0
6
4
Effective visual tool use hinges on causally grounded supervision, not just imitating tool calls, reshaping how we train multimodal agents.
DG-CFG not only enhances image generation quality but also reduces sampling costs by achieving target metrics with fewer steps, challenging the conventional wisdom of fixed guidance schedules.
Integrating 3D geometry, topological grammar, and physicochemical descriptors leads to a 20.6% reduction in prediction error for molecular properties, all while using under one million parameters.
OpenMobile proves that high-performing mobile agents can be trained on entirely synthetic, open-source data, closing the gap with closed-source models and enabling broader research.