Introduction to Deep Learning (1): Basics of Neural Networks

1. Deep learning concept

1. Definition

Deep learning solves feature engineering problems by training multi-layer network structures to classify or regress location data.

2. Deep learning applications

• Image Processing
• Language recognition
• natural language processing

It’s not very good on the mobile terminal. The amount of calculation is too large and the speed may be slow.
For example, medical applications and automatic coloring.

3. Example

Background dominance is the biggest problem when using k-nearest neighbors for judgment , because we focus on the subject (the main component)

2. Neural Network Basics

1. Linear function

For example, input a kitten picture through $f(x,W)$ to get the score for each category:

$f(x,$The weight parameter of each pixel in$W$$)$
$10$ : 10 categories
$b$ : Offset term (fine-tuning operation)
$w$ : weight parameter

2. Loss function

Regularization penalty term: Reduce the importance of a certain and prominent feature. Loss
function = data loss + regularization penalty term

3. Softmax classifier

We focus on the probability value of belonging to the correct category. The closer it is to 1, the smaller the loss, so the logarithmic function is used to calculate the loss value.

$ex$p：$e^x$ power, used to enlarge the mapping
$normalize$ : normalized
$L_i$: Actual loss value