Practical examples of Concepts of Neural Networks
汇报人：
11 月24日
神经网络概念的实用示例
思博
Sebastian Tabares
Purpose
Try to understand some notions of concepts leveraging graphical tools and interactions.
Content
- 1 neuron and problems that solve
- Activation Function
- 2 groups
- XOR
- 1 layer and problems that solve
- Features
- Neural networks
-
- Learning Rate
- Underfiting & Overfiting
- Regularization
- Train & Test data
- Real Data
Classification Problem (simplification)
1 variable
2 variable
2 categories
Activation function (purpose)
Lineal classification superposition create lineal results
This is the necessity to use "another element" in the architecture
Activation function (3D)
Activation function (3D)
1 Neuron
1 var, 2 vars, XOR
(perceptron)
1 Variable
this problem is easy for only one neuron
2 Variables
using 2 linear features, this problem can be solved easily
XOR Distribution (no linearly separable)
only a neuron can't solve this
1 layer
features
More complex example (lineal)
No*
This can be solved with 2 neurons?
*(using lineal features)More complex example (+f quadratic)
Using non-lineal features, you can solve it!*
*some times don't solve, so try again!More complex example (+f trigonometric)
Even is easier for trigonometric features!*
*only a neuron can solve this? try it!network
Underfitting, Overfitting, learning rate
Learning rate
Learning rate
The ability to deal with overfitting is what separates professionals and amateurs in machine learning.
处理过拟合的能力是区分机器学习专业人员和业余爱好者的关键所在
Overfitting
Underfitting
Overfitting
Regularization L1 & L2
Real Data
Thanks 谢谢
