Search papers, labs, and topics across Lattice.
Affiliation:
8
0
7
16
Action expert attention, not visual drift, is the true bottleneck for efficient KV caching in World Action Models, unlocking significant speedups.
Chromatin accessibility data, when properly integrated, unlocks significantly more accurate predictions of how cells will respond to genetic perturbations, even in entirely new cellular states.
Predictive pretraining for parametric PDEs unlocks a 51.4% generalization improvement on unseen governing parameters by explicitly modeling latent dynamics geometry.
LayerRecall boosts long-range video generation consistency by intelligently routing historical context to specific model layers, outperforming traditional methods without sacrificing local detail.
Bypassing RGB entirely, Latent-to-4D achieves significant improvements in 4D scene generation while maintaining a reusable framework across different video models.
SpecLoR cuts through noise and drift in text-to-video generation, achieving remarkable motion coherence with minimal computational cost.
By unifying geometric guidance across representation, architecture, and loss, UniGeo achieves unprecedented camera control and cross-view consistency in image editing, leaving previous fragmented approaches in the dust.
Advanced image editing models may look good but often miss the mark on logical consistency, revealing a critical gap in current AI capabilities.
