The Simplest Introduction to Deep Learning without any Mathematical Equation
Deep Learning is one type of Machine Learning technique that allow AI to learn without explicitly program which is just a mathematical equation represent in the form of Artificial Neural Network inspired from Neural Network in human brain.
Some Definitions
AI (Artificial Intelligence) is the computer program that can perform task that usually required human intelligence. ML (Machine learning) is a subset of AI which is the ability that allow AI to learn by itself without explicitly program. DL (Deep Learning) is a subset of ML which is still the ability that allow AI to learn by itself without explicitly program. However DL, unlike ML, uses ANN (Artificial Neural Network) as an infrastructure for learning process.
What inspired Deep Learning?
DL try to mimic the mechanism of the real Neural Network inside human brain which is the most powerful learning tool on the planet.
Neural Network inside human brain is made up of interconnection of many Neurons where each Neuron receive signal as an input from previous neuron(s), do some modification to the signal, and pass the signal as an output to another neuron(s) in order to make some decision about something.
The learning process occur when the Neurons adjust how they modify their input signal before passing it to the next Neuron(s) which change the influence of this neuron to another neuron, this affect the final decision made by human.
How Deep Learning mimics such mechanism?
In order for DL to mimic this mechanism, AN (Artificial Neuron) is invented to behave similarly like the Neuron in human brain.
However, this Artificial Neuron act as mathematical function that receive the input signal from previous Artificial Neuron in the form of matrix, do some mathematical modification to the input, and output the signal to the next Artificial Neuron in the form of matrix as well.
The interconnection of many Artificial Neurons construct the Artificial Neural Network where each interconnection contain a weight which determine the impact or importance of the signal output from the previous Artificial Neuron to the next Artificial Neuron. The process of passing the signal through Artificial Neural Network in order to generate some decision about something, prediction or classification, is called Forward Propagation. Where the name Deep Learning came from the depth of the complex interconnection of many Artificial Neurons to form the Artificial Neural Network. The more complex the ANN, the more complex function DL can estimate.
The intuition of learning process in Artificial Neural Network
There are mainly 2 processes in the learning process of Artificial Neural Network. First is the adjustment of the weight of the connection between 2 Artificial Neurons, this is to adjust how importance of the signal output from one Artificial Neuron to the another Artificial Neuron. Another is the adjustment of bias which transform the received input by doing some simple matrix addition or subtraction and pass it as the output for the next Artificial Neuron. The learning process use some optimization technique to minimize or maximize some objective function by varying the value of weight and the bias in Artificial Neural Network. The process of adjusting the weight and bias according to the observed objective function is called Backward Propagation.
Summary
Deep Learning is one type of machine learning technique that allow AI to learn without explicitly program which is just a complex set of many mathematical equation that receive some input and produce some output in terms of matrix that is represent in the form of Artificial Neural Network inspired from Neural Network in human brain.
