爱可可 AI 前沿推介(11.27)

文章预览

LG - 机器学习 CV - 计算机视觉 CL - 计算与语言 AS - 音频与语音 RO - 机器人 1、[LG] Cautious Optimizers：Improving Training with One Line of Code 2、[CL] Do Large Language Models Perform Latent Multi-Hop Reasoning without Exploiting Shortcuts? 3、[CL] From Jack of All Trades to Master of One：Specializing LLM-based Autoraters to a Test Set 4、[CL] Self-Generated Critiques Boost Reward Modeling for Language Models 5、[CL] Exploring Facets of Language Generation in the Limit 摘要：谨慎优化器、大型语言模型不利用捷径的潜多跳推理能力评估、基于大型语言模型的自动评估指标构建方法、用自生成批评促进语言模型的奖励建模、在极限下探索语言生成的各个方面 1、[LG] Cautious Optimizers: Improving Training with One Line of Code K Liang, L Chen, B Liu, Q Liu (2024) 谨慎优化器：用一行代码改进训练 要点: 一行代码的改进：本文提出了一种对任何基于动量的优化器 ………………………………