01_deeplearning_neural_networks_intuition

neural networks intuition

**origins: Algorithms that try to mimic the brain. **

Used in the 1980's and early 1990's. Fell out of favor in the late 1990's.

Fell out of favor in the late 1990's.

Resurgence from around 2005.

application : speech -> images -> text(NLP)

Biological neuron vs Simplified mathematical model of a neuron

dendrites:树突

axon:轴突

the performance of algorithms

we can control the number of large neural networks, so when we deal with the big data, in principle, we can just finish whatever data.

Demand Prediction

input : x: price

output: activation \(a = f(x) =\frac{1}{1+e^{-(wx+b)}}\)

so we can use a simple neuron to express it:

muti-layer demand prediction:

when the input is a vector: \(\vec{x}\): {price, shipping cost, marketing, material} then we get three parameters to consider :{affordability, awareness, perceived quality} and in the end we get the probability of being a top seller.

a layer is a grouping of neurons which have the same characteristics.

and layer can have multiple neurons

the outputs of middle layer are called activations.

to simplify, we just let all the inputs to function on the middle layer:

then: the middle layer that we also call it "hidden layer"

multiple hidden layers

we need to consider how many hidden layers and neurons

Face recognition

for example we get a vector: \(\vec{x}\)

then we can know that middle layers can deal with different windows of the picture.

the first middle layer can identify little regions of the picture, and the second middle layer is larger, and the third is more larger.

activations are higher level features

we get the probability of car detected.

in my opinion, we just make the problems to solve by many procedures.