blob: d8f90c39b47d74aec69911f2a7b6ce1ec5d38288 (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
|
%acmart
%IEEEtran
%rbt-mathnotes-formula-sheet
\documentclass{rbt-mathnotes-formula-sheet}
\usepackage[utf8]{inputenc}
\usepackage[pdf]{graphviz}
\usepackage{derivative}
\title{Deep learning notes}
\begin{document}
\section{Observations}
\begin{eqnarray}
i,j,k,l,L,m,M,n,N,o \in & \mathcal{N} \\
X \in & \mathcal{R}^{n \times o} \\
Y \in & \mathcal{R}^{n \times m}
\end{eqnarray}
\section{Neural Network}
\includegraphics[width=0.3\textwidth]{net.pdf}
\begin{eqnarray}
a^0 = & x_{1 \times p}(n) \\
a^L = & d_{1 \times m}(n) \\
a^l = & \varphi (z^l) \\
z^l = & a^{l - 1} W^l
\end{eqnarray}
\section{Gradient Descent}
\begin{eqnarray}
e(n) = & y(n) - d(n) \\
\xi(n) = & \frac{1}{2} e e^{\top}\\
\xi(n) = & \frac{1}{2} \sum_{j=1}^{M} (e_j(n))^2 \\
W_{(k + 1)} = & W_{(k)} - \nabla_{W} \xi(d,y) \\
\xi_{avg}(n) = & \frac{1}{2n} \sum_{n=1}^N \sum_{j=1}^{M} (e_j(n))^2 \\
\end{eqnarray}
\section{Backpropagation}
\begin{eqnarray}
\pdv{\xi}{\omega^l_{ij}} = & \delta_j^l \pdv{z_j^l}{\omega_{ij}} \\
\delta_j^l = & \pdv{\xi}{z_j^l} \\
\pdv{z_j^l}{\omega_{ij}} = & a_i^{l-1} \\
\pdv{\xi}{\omega_{ij}} = & \delta^l_{j} a_i^{l-1} \\
\end{eqnarray}
Output Layer
\begin{eqnarray}
\delta_j^L =& \pdv{\xi}{z_j^L} = \pdv{\xi}{a_j^L} \pdv{a_j^L}{z_j^L}\\
\delta_j^L =& \pdv{\xi}{a_j^L} \dot{\varphi}(z_j^L)\\
=& - e_j \dot{\varphi}(z_j^L)
\end{eqnarray}
Hidden Layer
\begin{eqnarray}
\delta_j^l = & \pdv{\xi}{z_j^l} = \sum_k \pdv{\xi}{z_k^{l+1}} \pdv{z_k^{l+1}}{z_j^l}\\
\delta_j^l = & \sum_k \delta_k^{l+1} \pdv{z_k^{l+1}}{z_j^l}\\
\pdv{z_k^{l+1}}{z_j^l} = &
\frac{\partial}{\partial z_j^l} \left( \sum_j \omega_{jk}^{l+1} \varphi(z_j^l) \right)\\
\pdv{z_k^{l+1}}{z_j^l} = & \omega_{jk} \dot{\varphi}(z_j^l)\\
\delta_j^l = & \sum_k \delta_k^{l+1} \omega_{jk}^{l+1} \dot{\varphi}(z_j^l)\\
\end{eqnarray}
\end{document}
|
