owned this note   changed 6 years ago  
Published Linked with GitHub

探索 Swift 自動微分實作 - John Lin

所有議程請見議程表。

歡迎來到 https://hackmd.io/@iPlayground/2019 共筆

Image Not Showing Possible Reasons
  • The image file may be corrupted
  • The server hosting the image is unavailable
  • The image path is incorrect
  • The image format is not supported
Learn More →

點擊本頁上方的 開始用 Markdown 一起寫筆記！
手機版請點選上方 按鈕展開議程列表。

請從這裡開始

YouTube

Image Not Showing Possible Reasons
  • The image file may be corrupted
  • The server hosting the image is unavailable
  • The image path is incorrect
  • The image format is not supported
Learn More →

  • Swift for TensorFlow 目標是Deap Laerning 讓寫AI 變簡單

  • 什麼是AI -> 圖像辨識

    • Neural Network , 模擬人腦的運作方式, Real Neuron (神經網路)

    • Artificial Neuron 加總向量

    • 將 Artificial Neuron 串起來，結合成 Train Network

    • 預測給定的Input 是屬於哪一個分類

How to know which Direction to move for the weight?

​​​​Neural network 實際上有兩種input (weight, training data) 的巨大函數

Linear regression with AI

var weight: SIMD<Float> = .random(in -1...1)
  • 透過 cost function 的微分，尋找谷底，也就是區域的最佳解。
  • 查找 wiki 時，可以選擇 "simple English" 😛

Main.SWift -> AST -> SILGen -> LLVM IR -> machine code

Parser -> Type Checker -> SILGen -> Mandatory Passes -> Opimization Passes -> IRGen -> LLVM







Image Not Showing Possible Reasons
  • The image file may be corrupted
  • The server hosting the image is unavailable
  • The image path is incorrect
  • The image format is not supported
Learn More →












結論:
Swift 將會內建自動微分功能

Resources:

  • Swift for Tensorflow
  • 已經有 mega proposal 要把它整合進一般的 swift， 未來在可能在 iOS 上也能用
tags:iPlayground2019
Select a repo