| generated by https://github.com/PolunLin/Udemy-copy-chapter
0. Introduction 2 堂講座 • 17 分鐘
0.0. Using Udemy like a pro 09:25
1. Download all course materials 2 堂講座 • 8 分鐘
1.0. Downloading and using the code 06:29
1.1. My policy on code-sharing 01:38
2. Concepts in deep learning 5 堂講座 • 1 小時 16 分鐘
2.0. What is an artificial neural network? 16:02
2.1. How models "learn" 12:26
2.2. The role of DL in science and knowledge 16:43
2.3. Are artificial "neurons" like biological neurons? 13:03
3. About the Python tutorial 1 堂講座 • 4 分鐘
3.0. Should you watch the Python tutorial? 04:25
4. Math, numpy, PyTorch 19 堂講座 • 3 小時 21 分鐘
4.0. PyTorch or TensorFlow? 00:44
4.1. Introduction to this section 02:06
4.2. Spectral theories in mathematics 09:16
4.3. Terms and datatypes in math and computers 07:05
4.4. Converting reality to numbers 06:33
4.5. Vector and matrix transpose 06:58
4.6. OMG it's the dot product! 09:45
4.7. Matrix multiplication 15:27
4.8. Softmax 19:26
4.9. Logarithms 08:26
4.10. Entropy and cross-entropy 18:18
4.11. Min/max and argmin/argmax 12:47
4.12. Mean and variance 15:34
4.13. Random sampling and sampling variability 11:18
4.14. Reproducible randomness via seeding 08:37
4.15. The t-test 13:57
4.16. Derivatives: intuition and polynomials 16:39
4.17. Derivatives find minima 08:32
4.18. Derivatives: product and chain rules 10:00
5. Gradient descent 10 堂講座 • 1 小時 57 分鐘
5.0. Overview of gradient descent 14:15
5.1. What about local minima? 11:56
5.2. Gradient descent in 1D 17:11
5.3. CodeChallenge: unfortunate starting value 11:30
5.4. CodeChallenge: 2D gradient ascent 14:48
5.5. Parametric experiments on g.d. 05:16
5.6. CodeChallenge: fixed vs. dynamic learning rate 18:56
5.7. Vanishing and exploding gradients 15:33
5.8. Tangent: Notebook revision history 06:04
6. ANNs (Artificial Neural Networks) 21 堂講座 • 5 小時 11 分鐘
6.0. The perceptron and ANN architecture 19:50
6.1. A geometric view of ANNs 13:38
6.2. ANN math part 1 (forward prop) 16:22
6.3. ANN math part 2 (errors, loss, cost) 10:54
6.4. ANN math part 3 (backprop) 12:10
6.5. ANN for regression 24:09
6.6. CodeChallenge: manipulate regression slopes 18:58
6.7. ANN for classifying qwerties 22:23
6.8. Multilayer ANN 23:46
6.9. Linear solutions to linear problems 19:51
6.10. Why multilayer linear models don't exist 08:14
6.11. Multi-output ANN (iris dataset) 06:20
6.12. CodeChallenge: more qwerties! 25:54
6.13. Comparing the number of hidden units 11:56
6.14. Depth vs. breadth: number of parameters 09:59
6.15. Defining models using sequential vs. class 17:25
6.16. Model depth vs. breadth 13:17
6.17. CodeChallenge: convert sequential to class 20:31
6.18. Diversity of ANN visual representations 06:37
6.19. Reflection: Are DL models understandable yet? 00:18
7. Overfitting and cross-validation 8 堂講座 • 1 小時 48 分鐘
7.0. What is overfitting and is it as bad as they say? 12:28
7.1. Cross-validation 17:13
7.2. Generalization 06:09
7.3. Cross-validation -- manual separation 12:39
7.4. Cross-validation -- scikitlearn 21:01
7.5. Cross-validation -- DataLoader 20:27
7.6. Splitting data into train, devset, test 09:45
7.7. Cross-validation on regression 08:09
8. Regularization 12 堂講座 • 2 小時 38 分鐘
8.0. Regularization: Concept and methods 13:38
8.1. train() and eval() modes 07:14
8.2. Dropout regularization 21:56
8.3. Dropout regularization in practice 23:13
8.4. Dropout example 2 06:33
8.5. Weight regularization (L1/L2): math 18:25
8.6. L2 regularization in practice 13:24
8.7. L1 regularization in practice 12:22
8.8. Training in mini-batches 11:32
8.9. Batch training in action 10:47
8.10. The importance of equal batch sizes 06:59
8.11. CodeChallenge: Effects of mini-batch size 11:57
9. Metaparameters (activations, optimizers) 24 堂講座 • 4 小時 52 分鐘
9.0. What are "metaparameters"? 05:02
9.1. The "wine quality" dataset 17:29
9.2. CodeChallenge: Minibatch size in the wine dataset 15:38
9.3. Data normalization 13:12
9.4. The importance of data normalization 09:33
9.5. Batch normalization 13:16
9.6. Batch normalization in practice 07:38
9.7. CodeChallenge: Batch-normalize the qwerties 05:06
9.8. Activation functions 17:59
9.9. Activation functions in PyTorch 12:12
9.10. Activation functions comparison 09:27
9.11. CodeChallenge: Predict sugar 07:48
9.12. Loss functions 17:06
9.13. Loss functions in PyTorch 16:50
9.14. More practice with multioutput ANNs 18:41
9.15. Optimizers (minibatch, momentum) 14:05
9.16. SGD with momentum 18:41
9.17. Optimizers (RMSprop, Adam) 07:46
9.18. Optimizers comparison 15:40
9.19. CodeChallenge: Optimizers and... something 10:17
9.20. CodeChallenge: Adam with L2 regularization 06:57
9.21. Learning rate decay 07:42
9.22. How to pick the right metaparameters 12:15
10. FFNs (Feed-Forward Networks) 12 堂講座 • 2 小時 15 分鐘
10.0. What are fully-connected and feedforward networks? 04:57
10.1. The MNIST dataset 12:33
10.2. FFN to classify digits 22:20
10.3. CodeChallenge: Binarized MNIST images 05:24
10.4. CodeChallenge: Data normalization 16:16
10.5. Distributions of weights pre- and post-learning 14:48
10.6. CodeChallenge: MNIST and breadth vs. depth 12:35
10.7. CodeChallenge: Optimizers and MNIST 07:06
10.8. Shifted MNIST 08:00
10.9. CodeChallenge: The mystery of the missing 7 11:25
10.10. Universal approximation theorem 10:47
11. More on data 11 堂講座 • 2 小時 25 分鐘
11.0. Anatomy of a torch dataset and dataloader 17:57
11.1. Data size and network size 16:35
11.2. CodeChallenge: unbalanced data 20:05
11.3. What to do about unbalanced designs? 07:45
11.4. Data oversampling in MNIST 16:30
11.5. Data noise augmentation (with devset+test) 13:16
11.6. Data feature augmentation 19:40
11.7. Getting data into colab 06:05
11.8. Save and load trained models 06:14
11.9. Save the best-performing model 15:18
11.10. Where to find online datasets 05:32
12. Measuring model performance 8 堂講座 • 1 小時 20 分鐘
12.0. Two perspectives of the world 07:01
12.1. Accuracy, precision, recall, F1 12:39
12.2. APRF in code 06:42
12.3. APRF example 1: wine quality 13:34
12.4. APRF example 2: MNIST 12:01
12.5. CodeChallenge: MNIST with unequal groups 09:14
12.6. Computation time 09:55
12.7. Better performance in test than train? 08:35
13. FFN milestone projects 6 堂講座 • 1 小時 2 分鐘
13.0. Project 1: A gratuitously complex adding machine 07:05
13.1. Project 1: My solution 11:18
13.2. Project 2: Predicting heart disease 07:14
13.3. Project 3: FFN for missing data interpolation 18:21
13.4. Project 3: My solution 09:35
14. Weight inits and investigations 10 堂講座 • 2 小時 19 分鐘
14.0. Explanation of weight matrix sizes 11:54
14.1. A surprising demo of weight initializations 15:52
14.2. Theory: Why and how to initialize weights 12:46
14.3. CodeChallenge: Weight variance inits 13:14
14.4. Xavier and Kaiming initializations 15:42
14.5. CodeChallenge: Xavier vs. Kaiming 16:54
14.6. CodeChallenge: Identically random weights 12:40
14.7. Freezing weights during learning 12:58
14.8. Learning-related changes in weights 21:55
14.9. Use default inits or apply your own? 04:36
15. Autoencoders 6 堂講座 • 1 小時 51 分鐘
15.0. What are autoencoders and what do they do? 11:42
15.1. Denoising MNIST 15:48
15.2. CodeChallenge: How many units? 19:52
15.3. The latent code of MNIST 17:55
15.4. Autoencoder with tied weights 21:57
16. Running models on a GPU 3 堂講座 • 32 分鐘
16.0. What is a GPU and why use it? 15:07
16.1. Implementation 10:13
16.2. CodeChallenge: Run an experiment on the GPU 06:46
17. Convolution and transformations 12 堂講座 • 2 小時 56 分鐘
17.0. Convolution: concepts 21:33
17.1. Feature maps and convolution kernels 09:32
17.2. Convolution in code 21:05
17.3. Convolution parameters (stride, padding) 12:14
17.4. The Conv2 class in PyTorch 13:23
17.5. CodeChallenge: Choose the parameters 07:10
17.6. Transpose convolution 13:41
17.7. Max/mean pooling 18:35
17.8. Pooling in PyTorch 13:29
17.9. To pool or to stride? 09:35
17.10. Image transforms 16:47
17.11. Creating and using custom DataLoaders 19:06
18. Understand and design CNNs 16 堂講座 • 4 小時 10 分鐘
18.0. The canonical CNN architecture 10:47
18.1. CNN to classify MNIST digits 26:06
18.2. CNN on shifted MNIST 08:36
18.3. Classify Gaussian blurs 24:10
18.4. Examine feature map activations 27:50
18.5. CodeChallenge: Softcode internal parameters 16:48
18.6. CodeChallenge: How wide the FC? 11:25
18.7. Do autoencoders clean Gaussians? 17:10
18.8. CodeChallenge: AEs and occluded Gaussians 09:36
18.9. CodeChallenge: Custom loss functions 20:15
18.10. The EMNIST dataset (letter recognition) 16:59
18.11. Dropout in CNNs 24:59
18.12. CodeChallenge: How low can you go? 10:14
18.13. CodeChallenge: Varying number of channels 06:45
18.14. So many possibilities! How to create a CNN? 13:39
19. CNN milestone projects 5 堂講座 • 40 分鐘
19.0. Project 1: Import and classify CIFAR10 07:15
19.1. Project 1: My solution 12:01
19.2. Project 2: CIFAR-autoencoder 04:51
19.3. Project 3: FMNIST 03:52
20. Transfer learning 8 堂講座 • 1 小時 46 分鐘
20.0. Transfer learning: What, why, and when? 16:52
20.1. Transfer learning: MNIST -> FMNIST 10:06
20.2. CodeChallenge: letters to numbers 14:06
20.3. Famous CNN architectures 06:46
20.4. Transfer learning with ResNet-18 16:43
20.5. Pretraining with autoencoders 03:41
20.6. CIFAR10 with autoencoder-pretrained model 20:01
21. Style transfer 5 堂講座 • 58 分鐘
21.0. What is style transfer and how does it work? 04:36
21.1. The style transfer algorithm 12:37
21.2. Transferring the screaming bathtub 10:58
21.3. CodeChallenge: Style transfer with AlexNet 22:16
22. Generative adversarial networks 7 堂講座 • 1 小時 25 分鐘
22.0. GAN: What, why, and how 17:22
22.1. Linear GAN with MNIST 21:55
22.2. CodeChallenge: Linear GAN with FMNIST 09:50
22.3. CNN GAN with Gaussians 15:06
22.4. CodeChallenge: Gaussians with fewer layers 06:05
22.5. CNN GAN with FMNIST 06:24
22.6. CodeChallenge: CNN GAN with CIFAR 07:51
23. RNNs (Recurrent Neural Networks) (and GRU/LSTM) 9 堂講座 • 2 小時 48 分鐘
23.0. Leveraging sequences in deep learning 12:53
23.1. How RNNs work 15:14
23.2. The RNN class in PyTorch 17:44
23.3. Predicting alternating sequences 19:30
23.4. CodeChallenge: sine wave extrapolation 24:49
23.5. GRU and LSTM 15:51
23.6. The LSTM and GRU classes 23:08
23.7. Lorem ipsum 13:26
24. Ethics of deep learning 5 堂講座 • 49 分鐘
24.0. Will AI save us or destroy us? 09:40
24.1. Example case studies 06:39
24.2. Some other possible ethical scenarios 10:35
24.3. Will deep learning take our jobs? 10:27
24.4. Accountability and making ethical AI 11:22
25. Where to go from here? 2 堂講座 • 24 分鐘
25.0. How to learn topic _X_ in deep learning? 08:08
25.1. How to read academic DL papers 16:00
26. Python intro: Data types 8 堂講座 • 1 小時 41 分鐘
26.0. How to learn from the Python tutorial 03:25
26.1. Variables 18:14
26.2. Math and printing 18:31
26.3. Lists (1 of 2) 13:31
26.4. Lists (2 of 2) 09:29
26.5. Tuples 07:40
26.6. Booleans 18:19
26.7. Dictionaries 11:51
27. Python intro: Indexing, slicing 2 堂講座 • 24 分鐘
27.0. Indexing 12:30
27.1. Slicing 11:45
28. Python intro: Functions 8 堂講座 • 1 小時 41 分鐘
28.0. Inputs and outputs 07:01
28.1. Python libraries (numpy) 14:20
28.2. Python libraries (pandas) 13:57
28.3. Getting help on functions 07:36
28.4. Creating functions 20:27
28.5. Global and local variable scopes 13:20
28.6. Copies and referents of variables 05:45
28.7. Classes and object-oriented programming 18:46
29. Python intro: Flow control 10 堂講座 • 2 小時 36 分鐘
29.0. If-else statements 15:03
29.1. If-else statements, part 2 16:58
29.2. For loops 17:37
29.3. Continue 12:11
29.4. Initializing variables 07:24
29.5. Single-line loops (list comprehension) 18:01
29.6. while loops 15:25
29.7. Broadcasting in numpy 19:30
29.8. Function error checking and handling 15:41
30. Python intro: Text and plots 7 堂講座 • 1 小時 42 分鐘
30.0. Printing and string interpolation 17:18
30.1. Plotting dots and lines 12:55
30.2. Subplot geometry 16:10
30.3. Making the graphs look nicer 18:48
30.4. Seaborn 11:08
30.5. Images 17:59
30.6. Export plots in low and high resolution 07:58
31. Bonus section 1 堂講座 • 1 分鐘
31.0. Bonus content 00:53
===
===
- add number in folder name
- add number in file name
- add md file in each folder