清華大學應數系
演算法
Algorithms
Robotic AI Lab
Applied Math, National Tsing Hua
University
劉晉良
Jinn-Liang Liu
2019 Spring Course
演算法 Algorithms
南大校區9307,
1:20 – 2:50 – 4:10pm, Mondays (2019.2.22 – 2019.6.10)
Reports: Do Proj1
and a project of your choice and Send reports in ppt (Demo) to jinnliu@mail.nd.nthu.edu.tw
(email 標題:姓名 學號 Report 1, 2, or
3) by
Mondays 3/25; 4/29;
6/10.
工業革命:動力1.0 (1760),電力2.0 (1870),數位3.0 (1945),智力4.0 (2016 AlphaGo)
AI 聽說讀寫食衣住行育樂醫金…, ABC.
l AI abc: An Introduction to Machine Learning
l Gradient Descent and Backpropagation in Machine Learning (Automatic Differentiation:
Forward & Reverse Modes, Jacobian)
l Convolution in
Machine Learning (Convolution)
Part I Computer
Programming (Browse and Use) GitHub
2.
TensorFlow, TensorBoard
(YouTube1),
Colab: 1. Style
Transfer (YouTube2, Code2),
2. Time Series (YouTube3), 3. YOLO (YouTube4, Code3),
4. Stock (YouTube5, Code4),
5. Game (OpenAI
RL Code5)
(Proj1: A. Colab, tf3-MNIST.py, keras_tensorboard.ipynb
(Tutorial))
3.
PyTorch (PyTorch
Autograd, PyTorch入門,
MNIST.ipynb, MNIST.py)
A. Cloud Computing by Colab: Google Chrome => Google Account => Google 雲端硬碟 => Click tf1.ipynb => Upload tf1.ipynb
to Google 雲端硬碟 => Click on tf1.ipynb => Install Colab Notebooks => Run tf1.ipynb
(Done!)
B. Local Computing by jupyter: Install Anaconda3-4.2.0 (or more Anaconda) => Anaconda Navigator => Environments => Install TensorFlow => Home => Launch jupyter => jupyter => Files
on 筆電 => Click on tf1.ipynb => Run tf1.ipynb (Done!)
How to run py code on jupyter:
tf1.py => Creat a new file tf1.ipynb with only one line
“import tf1” => Run the cell of “import tf1” (Done!)
C.
Install PyTorch
via Anaconda: Click Anaconda Prompt => >conda install PyTorch -c PyTorch => (take a while) => >pip install torchvision => Try 1.
import numpy as np 2. import
torch 3. x = torch.empty(5, 3) 4. print(x) on jupyter (Done!)
D. Local Computing by PyCharm:
1. tf1.py: # Put this file tf1.py in D:\AI\TF (in English, no Chinese).
# Install Anaconda3-4.2.0 => Install PyCharm Community Edition 2017.1.5 for Windows
=> PyCharm => Create New Project => Location: D:\AI\TF
=> Interpreter: Click on Create Conda Env (Far Right Button)
=> Click Create => Click File => Open tf1.py => Tools
=> Python Console => >>> import pip => >>> pip.main(['install', 'tensorflow'])
=> Wait for >>> => Click the green triangle button (Run tf1). Done!
# See YouTube Link for more if needed.
2. tf2.py: # How to use TensorBoard. # PyCharm => View => Tool Windows
=> Terminal => Click the green + sign in the Terminal Screen
=> (C:\Users\Jinn\Anaconda3\envs\untitled1) D:\AI\TF> tensorboard --logdir="./graphs" --port 6006
=> Copy http://Jinn-PC:6006 => Google Chrome
=> Paste http://Jinn-PC:6006 to Chrome's http address
=> GRAPHS (the graph of a, b, Add)
3. TF mnist 1.0 (tf3-MNIST.py) => # Google Chrome => http://Jinn-PC:6006 => GRAPHS => # Error: matplotlib not installed
=> # PyCharm => File => Setting => Project: TF => Project Interpreter => + => matplotlib
=> Specify version => Install # Save zout_MNIST_1.0.png
Part II Supervised
Learning (Read and Work)
1.
A Simple Learning Model: Classification, Target,
Hypothesis, Training Data, Learning Algorithm, Weights, Bias, Supervised and
Unsupervised Learning
2.
Google Tutorial
for ML Beginners: Image Recognition, MNIST, Softmax
Regression (92%), Cross Entropy, Gradient Descent, Back Propagation,
Computation Graph (TF
mnist 1.0)
3.
Tensorflow
and Deep Learning I (by Martin Gorner) : Deep Learning Network, ReLU,
Learning Rate (98%), Overfitting, Dropout (98.2%), Convolutional Neural Network (99.3%) (TF
mnist 3.1)
4.
Tensorflow
and Deep Learning II (by Martin Gorner) (RNN1): Batch
Normalization (99.5%) (TF
mnist 4.2), Data Whitening,
Fully Connected Network, TensorFlow API, MNIST Record
(Kaggle: 100%), Recurrent Neural Network, Deep RNN, Long
Short Term Memory, Gated Recurrent Unit, Language Model
Part III Theories of Deep Learning
1.
Lectures at MIT
(Book: Deep
Learning by Goodfellow, Bengio, Courville)