华为云官方博客 ·

R-Drop论文复现与理论讲解

💡 原文中文，约7400字，阅读约需18分钟。

📝

内容提要

本文介绍了R-Drop，一种基于Dropout的正则化方法，它通过使用若干次Dropout，定义新的损失函数，约束两次输出一致，以防止模型过拟合，提高模型正确率。实验结果表明，R-Drop能够明显提升模型性能。

🏷️

刚刚，Hinton发了首篇RSI论文
辛顿等22位学者发表论文探讨AI递归自我改进。Anthropic数据显示，AI自主研发比例半年增长二十余倍，AI正加速进入AI研发流水线。AI研究员可复制...
NEAREST BY Join: Scaling Vector Search in Databricks Runtime
Vector search originated as a serving problem. The classical use case is a chatbot...
The Future of Mathematics
[This is a guest post by Jeremy Avigad. This blog post was initially written ...
使用Markdown和Docbook编写Perl文档（包括POD）
作者发布了多个CPAN模块，支持用Markdown或Docbook编写Perl文档，再合并为POD嵌入源码，并发布为静态HTML站点。其中Markdown...
Hyundai CEO says only a ‘level playing field’ can minimize damage from China
There's been a lot of doom and gloom from the auto industry lately when t...
在笔记本电脑上运行、几乎媲美35B模型的AI安全检查
Guardrail selection has typically meant choosing between a purpose-built clas...