Search papers, labs, and topics across Lattice.
Affiliation:
6
1
11
0
Subtle prompt changes can destabilize LLMs significantly, but four key factors can mitigate this sensitivity by targeting low-order interactions.
Planning performance can be dramatically improved by aligning latent embedding geometry with task-relevant state representations, as shown by SCALE's consistent outperformance of LeWM.
LASER cuts average latency by up to 38% while only sacrificing 1% accuracy, revolutionizing how we deploy reasoning models on edge devices.
RISE achieves a remarkable 2.1脳 speedup in text-to-image diffusion tasks by intelligently splitting the denoising workload between edge and device models without compromising quality.
By fusing sparse coding with visual Transformers, LISTA-Transformer achieves state-of-the-art fault diagnosis accuracy, surpassing traditional methods by 3.3% on the CWRU dataset.
HiLoRA achieves superior personalization in federated learning by adaptively clustering clients based on LoRA subspace similarity, enabling targeted knowledge sharing and outperforming standard LoRA-based methods.
