What is a Neural Network？
Easily，Deep Learning is a more complex Neural Network.
We can predict the house prices from the size.Based on it,We can fit a function to predict house prices.
But you may find it is strange that the prices can not be negative.So we should bend the straight line a little bit to let it finally end at zero.
This is almost certainly the simplest neural network.
In the literature on neural networks,you can usually find this function.It is called ReLU activation function(Rectified Linear Unit).
But we can't easily predict the price just from its area in reality.We need other features to help us reach the goal.
Each circle can be one part of ReLU. For the house , These are all things that go hand in hand with it. By stacking these individual neurons together,we obtain a slightly neural network.This shows the magic of neural network,although i have described it,it can also need area,degree of walkability ,quantity of nearby school and any function that can influence the price.
Supervived Learning with Neural Networks
In the suoervived Learning with Neural networks,you have some inputs(x),and you want to get a function to map some outputs(y).Like the housing price prediction we mentioned before,you can only input some features about the house,and try to get the output or estimate the price.
For image applications,we frequently employ convolutions on neural networks,commonly abbreviated as CNNs.For sequential data,such as audio,there is a temporal component,as the audio is played out over time,making it natural reperesentation as a one-dimensional time series or temporal sequence.Recurrent Neural Networks(RNNs)are often utilized for sequential data.Language, with English and Chinese alphabets or words appearing one after another, is also inherently sequential, making more sophisticated versions of RNNs frequently employed for these applications.
In more complex applications like autonomous driving, where an image may display more CNN (Convolutional Neural Network) architectures, the radar information is entirely different, potentially requiring a more customized or some hybrid neural network architecture. To more concretely illustrate what constitutes a standard CNN or RNN structure, you might have encountered images in the literature that depict a standard neural network.
You've probably also heard about the use of machine learning fot both structured and unstructured data.
Why is Deep Learning taking off
The main functions may be three parts:Data Scale,Computational,Algorithm.
Data Scale
Computational
CPU and GPU
Algorithm
ReLU->sigmoid
