AI & ML & DL

1102 機器學習

這篇文章是學習時整理的一些筆記，讓自己複習時方便，文章部分內容為上課之內容及閱讀清單之整理

Introduction

amd

AI 可以看作是一個總稱，泛指可以表現智慧的系統或是機器
ML 主要在研究如何建構能依據自身經驗自我改善的系統
而 DL 是 ML 的一個分支，是模擬人類神經元的人工神經網路為架構對資料進行特徵學習

ML 和 DL 最大的差別在於：ML是由人工挑選特徵並讓機器學習，而DL是由機器直接學習特徵與資料

amd

History

1950s Artificial Intelligence

Artificial Intelligence

Machine learning

1980 Artificial Neural Network
1986 Back Propagation is proposed to solve the complex computation in neural network(Gradient Vanish Problem)
Decision Tree, Forest Tree, Support Vector Machine, …becomes popular

2006 Neural network

Prof. Hinton utilized Restricted Boltzmann Machine to train neural network.
Bad impression of neural network. => Deep Learning
Decision Tree, Forest Tree, Support Vector Machine becomes shallow learning.

2007 ImageNet

Since the launch of the ImageNet competition in 2007, the results of the error rate is roughly 30%, 29%, 28% in each year’s competitions.

2012 AlextNet

Prof. Hinton used deep learning (AlextNet) to make error rate become 16.42%.
那時候造成很大的轟動，也使得CNN開始被重視，變成所謂的CNN大時代，之後的比賽也都是由CNN拿下冠軍，深度學習正式大爆發。