爱可可 AI 前沿推介(10.24)

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] Simplicity Bias via Global Convergence of Sharpness Minimization 2、[LG] xLSTM-Mixer：Multivariate Time Series Forecasting by Mixing via Scalar Memories 3、[LG] Building Conformal Prediction Intervals with Approximate Message Passing 4、[LG] Pyramid Vector Quantization for LLMs 5、[CL] Fine-Tuning Large Language Models to Appropriately Abstain with Semantic Entropy 摘要：锐度最小化与简洁性偏差、通过标量记忆混合进行多元时间序列预测、用近似消息传递构建保形预测区间、大语言模型的金字塔向量量化、基于语义熵的大型语言模型微调 1、[LG] Simplicity Bias via Global Convergence of Sharpness Minimization K Gatmiry, Z Li, S J. Reddi, S Jegelka [MIT & TTIC & Google Research] 锐度最小化与简洁性偏差 要点: 简洁性偏差与锐度最小化: 本文研究了随机梯度下降 (SGD) 的两种看似不同 ………………………………