Machine Learning - Day2 - Deep Learning Intro

Edible Instruction

1. Too many formulations and equations, so screen shots of PPT is necessary.
2. My mathematics sucks, so you 'll see many mathematical knowledge interspersed in my all ML learning session.
3. Don’t be afraid of any mathematical symbol, just BAIDU the definition.

linear models are too simple

We need more sophisticated models. (Red curve, but we can’t use linear model to create the shape of red curve)
Linear model have severe limitation. We need a more flexible model!

Red curve = constant (常数)+ sum of a set of threshold(阈值)
0 + 1 + 2 + 3 = Red curve

All Piecewise Linear Curves

= constant + sum of a set of threshold

More pieces require more threshold

Beyond Piecewise Linear?

To have good approximation, we need sufficient pieces.

How to represent this function?

Sigmoid function, refer to CSDN Article Baidu Article

Nature Exponential e (2.71828182845904523536) Baidu Article
c = constant

b, w and c are unknown parameters

We got the new model: More features

j: 1,2,3 means the number of features
$X_1$ : viewers from the previous day
$X_2$: viewer from two days ago
$X_3$: viewer from three days ago

i: each blue function

$r_1$, $r_2$, $r_3$: the results of

r is vector (What’s vector? Article)
W is matrix(What’s matrix? Article)
How to calculate matrix and vector? Article

$C^T$ T is transpose (Article)

Function with unknown parameters

Back to ML framework

Step1 completed

Move to Step2 and Step3

θ called theta, means all unknown parameters in here.
$\nabla$: called nabla,

epoch: see all the batches once
update: iteration

More variety of models …

Activation function (The best interpretation Article)

Even more variety of models …

Neural Network

Overfitting

Overfitting Article