[ML] Artificial Neural Networks (ANN, 인공 신경망)

• The gateway to reinforcement learning
• A popular machine learning technique for analyzing data through a network of decision layers.
• ANN consists of interconnected decision functions(nodes), which interact with each other through axon-like edges.
• The first layer consists of raw input data(numeric values, text, image pixels or sound) divided into nodes.
• Each input node sends information to the next layer of nodes via the network’s edges.

• Nodes,  edges/weights, sum/activation function of a basic neural network
• Each edge in the network has a numeric weight that can be altered based on experience.
• If the sum of the connectected edges satisfies a set threshold, known as the activation function, this activates a neuron at the next layer.
• If the sum of the connected edges does not meet the set threshold, the activation function fails, which results in an all or nothing arrangement.Using supervised learning, the model’s predicted output is compared to the actual output and the difference between these two results is measured as the cost or cost value.
  
• The purpose of training is to reduce the cost value until the model’s prediction closely matches the correct output.
• By incrementally tweaking the network’s weights
• This process of training the neural network is called “back-propagation”.

Black-Box Dilemma

• The model is unable to specify the relationship of these independent variables to the dependent variable of the problem reaching the target.
• Algorithms such as decision trees and linear regression are transparent as they show the variables’ relationships to a given output.

Building a Neural Network

• Input Layer: data is received and features are detected.
• Hidden Layer: analyze and process the input features.
• Output Layer: final result is shown.
   
• As more hidden layers are added to the network, the model’s capacity to analyze complex patterns also improves.
• This is why models with a deep number of layers are often referred to as “deep learning” to distinguish their deeper and superior processing abilities.
• The most basic form of a feed-forward neural network is the perceptron.

Perceptron builds a Prediction Model