Search papers, labs, and topics across Lattice.
Huazhong University of Science and Technology
5
0
6
AutoMine outperforms existing methods in scenario mining, achieving a HOTA-Temporal score of 36.38 in a competitive setting.
Self-supervised learning can transform low-quality labels into a powerful asset for enhancing underwater images, leading to unprecedented restoration quality.
MLLMs can be significantly improved by directly supervising visual tokens with corresponding text, without needing architectural changes or extra computation.
Latent reasoning can beat explicit Chain-of-Thought – but only if you force it to learn causal dynamics via a visual world model, not just language.
Naturalness-based data selection, a common technique for curating LLM reasoning datasets, systematically favors longer, lower-quality reasoning chains due to a previously unnoticed "step length confounding" effect.
