Search papers, labs, and topics across Lattice.
College of Computing and Data Science, Nanyang Technological University, Singapore
5
0
8
Jointly pruning dependent units in LLMs can lead to more effective model compression without the need for fine-tuning, challenging conventional independent pruning methods.
ARDepth reveals that structured auto-regressive generation can significantly enhance monocular depth estimation by capturing local details without sacrificing global coherence.
LLMs struggle with negated commonsense reasoning, but pre-training on automatically-generated negated knowledge triples can significantly improve their performance.
Seedance 2.0 leapfrogs existing models by unifying multi-modal inputs (text, image, audio, video) into a single architecture for generating high-quality, longer-duration audio-video content.
Agentic LLMs are far more vulnerable to indirect prompt injection attacks than previously thought: AdapTools achieves over 2x improvement in attack success while significantly degrading system utility, even against strong defenses.
