爱可可 AI 前沿推介(5.29)

文章预览

LG - 机器学习 CV - 计算机视觉 CL - 计算与语言 1、[LG] EM Distillation for One-step Diffusion Models 2、[CL] Can Large Language Models Faithfully Express Their Intrinsic Uncertainty in Words? 3、[AI] Devil's Advocate：Anticipatory Reflection for LLM Agents 4、[LG] Transformers Can Do Arithmetic with the Right Embeddings 5、[CL] NV-Embed：Improved Techniques for Training LLMs as Generalist Embedding Models 摘要：单步扩散模型的EM蒸馏、大型语言模型能否用语言忠实表达其内在不确定性、LLM Agent的预期反思、Transformer可以通过正确的嵌入进行算术运算、将LLM训练成通用嵌入模型的改进技术 1、[LG] EM Distillation for One-step Diffusion Models S Xie, Z Xiao, D P Kingma, T Hou… [Google DeepMind & Google Research] 单步扩散模型的EM蒸馏 要点: 虽然扩散模型可以生成高质量样本，但是需要耗费大量计算的迭代采样过程，现有的蒸馏方法在极少采样步数下效果降低 ………………………………