Search papers, labs, and topics across Lattice.
8
0
12
3
WorldFly's innovative approach to integrating world models enables UAVs to navigate complex urban landscapes with unprecedented robustness.
The hardest AI tasks remain largely unsolved, with current models achieving only a 2.6% success rate on economically valuable workflows.
Modeling high-order interactions among traffic nodes via prototype-guided hypergraph construction significantly boosts spatiotemporal forecasting accuracy.
Decomposing traffic signals into regular and fluctuating components, then applying separate temporal modules, dramatically improves traffic forecasting accuracy compared to treating traffic as a unified representation.
LLMs might understand causality better than their Yes/No answers suggest, meaning current benchmarks could be misleadingly pessimistic.
Compressing visual tokens doesn't have to mean sacrificing performance: VEN-VL's ensemble-MoE approach recovers accuracy while maintaining efficiency in multimodal models.
LLMs can reason more effectively by directly tracking their own belief in the correct answer throughout the reasoning process, enabling more targeted policy updates.
Achieve state-of-the-art autonomous UAV exploration performance with 87% less computation by using a hierarchical planner that only invokes global replanning when absolutely necessary.
