Search papers, labs, and topics across Lattice.
6
0
9
30
LLMs can learn to generate syntactically valid text without sacrificing semantic quality, even with underspecified prompts, by leveraging self-supervised rewards derived from their own likelihoods.
Most multimodal models fail to truly understand tool use in real-world egocentric videos, with even Gemini-3.1-Pro struggling to ground actions in visual evidence.
Synthetic data, when carefully aligned with real-world characteristics, can boost hand-object interaction detection by over 11% even when real labeled data is scarce.
Current methods struggle to understand human behavior in industrial settings, as evidenced by the challenging ENIGMA-360 dataset of synchronized ego-exo videos.
Differentiable Viterbi decoding unlocks state-of-the-art procedural planning with an order of magnitude fewer parameters than LLM-based approaches.
By modeling environment affordances and interaction hotspots, STAformer++ boosts short-term object interaction anticipation by up to 31 percentage points on EPIC-Kitchens.
