Search papers, labs, and topics across Lattice.
Beijing University of Posts and Telecommunications
6
5
12
LLMs are surprisingly bad at identifying security goals in protocol documents, but instruction tuning on a new benchmark, SecGoal, closes the gap.
LLM-generated rewards in RL can actually hurt performance if deployed at the wrong training stage, but this competence-aware verification method can help.
Skip the retraining: AM-SGHMC lets you apply a single trained MCMC sampler to various Bayesian updating problems for similar structures.
Diffusion models can now reason recursively over visual tokens, achieving state-of-the-art image generation performance by dynamically selecting specialized neural modules at each diffusion step.
A unified benchmark reveals the trade-offs between pixel-wise accuracy and perceptual realism in state-of-the-art image super-resolution techniques.
Don't let your SWE agent drown in context: SWE-AGILE maintains performance on multi-turn software engineering tasks by dynamically managing reasoning context with a novel sliding window and compressed reasoning digests.
