Search papers, labs, and topics across Lattice.
7
0
8
Video world models can now faithfully respect coarse, user-designed 3D scene proxies without ever training on a single hand-crafted proxy-video pair.
Dynamic MoE load balancing can run at every single microbatch with zero CPU synchronization by having GPUs deterministically solve bitwise-identical expert replication plans locally, boosting cluster throughput by up to 11.4%.
Treating 3D Gaussian reconstruction as an active structural prior rather than a passive auxiliary target drives a 22% performance leap in category-level 6D pose estimation.
Pallas cuts service interruption time by up to 89.68 times during LLM inference handovers, transforming mobile AI experiences.
PANY outperforms existing model-free methods by over 20% in pose accuracy, even in challenging conditions with limited reference overlap.
Training LLMs across regions doesn't have to break the bank: BACE-Pipe cuts job completion time by up to 65% and electricity costs by 30% by intelligently scheduling pipeline stages across different geographic locations.
Forget ImageNet: Xray-Visual sets a new SOTA for multimodal vision models by scaling to billions of social media data points with a novel three-stage training pipeline.
