跳到主要内容Skip to main content

神经网络Neural Network

Neural Network

  • 核心概念Core concepts
  • 第一站 · 进门Station 1 · Getting In
解决什么问题What problem it solves

模仿大脑神经元连接方式的一类数学结构,是现代 AI 的通用积木。A network of simple units whose combined output can approximate very complex functions.

这是什么What it is

由大量简单计算单元(神经元)分层连接、互相协作完成计算的结构。

每个单元做的事很简单(加权求和 + 激活),难的是整体如何调参数。

Layers of connected units, each applying a small transformation, whose overall behaviour is learned from data rather than specified.

为什么需要知道Why it matters

它是「看懂站」的基础设施。理解了它,后面的 Transformer 才好理解——那是它的一个变种。It is the mechanism behind deep learning. You do not need the maths to use it, but it explains both what it can do and why it fails in particular ways.

出处Sources
  • CS231n:神经网络(一)讲义CS231n: Neural Networks Part 1 (notes) ↗
  • 《深度学习》第6章:深度前馈网络Deep Learning, Ch.6: Deep Feedforward Networks ↗