Google Transformer V2:嵌套学习(Nested Learning)架构

Type: article
Author: Google Research
Primary Topic: 行业动态
Ingested: 2026-05-19

Summary

Google提出「嵌套学习(Nested Learning)」架构,被视为《Attention Is All You Need》的2.0版本。该架构将所有计算组件视为关联记忆模块,支持持续学习和自我参数修改,解决了传统Transformer静态权重、扩展边际递减等问题。基于此架构的Hope模块在持续学习任务和长文本推理(支持10M Token上下文)上显著优于现有架构。

Key Concepts

Entities

Source

Relations


Auto-generated on 2026-05-19