1 |
Neural Networks and Deep Learning |
text |
2 |
What this book is about |
text |
3 |
On the exercises and problems |
text |
4 |
- |
image |
5 |
Using neural nets to recognize handwritten digits |
text |
6 |
Perceptrons |
text |
7 |
Sigmoid neurons |
text |
8 |
The architecture of neural networks |
text |
9 |
A simple network to cla***ify handwritten digits |
text |
10 |
Learning with gradient descent |
text |
11 |
Implementing our network to cla***ify digits |
text |
12 |
Toward deep learning |
text |
13 |
- |
image |
14 |
How the backpropagation algorithm works |
text |
15 |
Warm up: a fast matrix-based approach to computing the output from a neural network |
text |
16 |
The two a***umptions we need about the cost function |
text |
17 |
The Hadamard product, $s \odot t$ |
text |
18 |
The four fundamental equations behind backpropagation |
text |
19 |
Proof of the four fundamental equations (optional) |
text |
20 |
The backpropagation algorithm |
text |
21 |
The code for backpropagation |
text |
22 |
In what sense is backpropagation a fast algorithm? |
text |
23 |
Backpropagation: the big picture |
text |
24 |
- |
image |
25 |
Improving the way neural networks learn |
text |
26 |
The cross-entropy cost function |
text |
27 |
Overf****ing and regularization |
text |
28 |
Weight initialization |
text |
29 |
Handwriting recognition revisited: the code |
text |
30 |
How to choose a neural network's hyper-parameters? |
text |
31 |
Other techniques |
text |
32 |
- |
image |
33 |
A visual proof that neural nets can compute any function |
text |
34 |
Two caveats |
text |
35 |
Universality with one input and one output |
text |
36 |
Many input variables |
text |
37 |
Extension beyond sigmoid neurons |
text |
38 |
Fixing up the step functions |
text |
39 |
Conclusion |
text |
40 |
- |
image |
41 |
Why are deep neural networks hard to train? |
text |
42 |
The vanis***ng gradient problem |
text |
43 |
What's causing the vanis***ng gradient problem? Unstable gradients in deep neural nets |
text |
44 |
Unstable gradients in more complex networks |
text |
45 |
Other obstacles to deep learning |
text |
46 |
- |
image |
47 |
Deep learning |
text |
48 |
Introducing convolutional networks |
text |
49 |
Convolutional neural networks in practice |
text |
50 |
The code for our convolutional networks |
text |