Neural Networks: You Don't Know 3 Features | Medical Pharma News

Artificial neural networks (ANNs) are a computer model that simulates the behavior of the human brain. Find out how.

Neural networks were a computer model created by Warren McCulloch and Walter Pitts in 1943, known as threshold logic, which is based on mathematics and algorithms.

Since this initiative, neural network research has divided into two distinct areas. One focused on biological processes in the brain, and the other was based on the application of neural networks to artificial intelligence.

Neural network models in artificial intelligence usually refer to artificial neural networks (ANNs); they are basic mathematical models that define a function f:X→Y or a distribution plus X or both X and Y.

The term “network” refers to the interconnections between neurons in the different layers of each system.

How does it work?

An artificial neural network is made up of artificial neurons, which receive information from the outside or from other neurons, process it, and generate an output value that is fed to other neurons in the network or that is the response or output to the outside of the network.

An ANN is defined by the following characteristics:

The interconnection model between the different layers of neurons.
The activation function that converts the weighted inputs of a neuron into its activation output.
The learning phase for updating the interconnection weights.

What has captivated interest in neural networks is their learning potential. In this case, given a task to be solved and a class of functions F, learning consists of using a set of observations to find f∗ ∈ F that solves the task well.

Most of the algorithms used in the interconnections between the artificial neurons used for training employ some form of gradient descent.

Advantage of ANN

The greatest advantage of ANNs is their ability to be used as an arbitrary function approximation mechanism that “learns” from observed data. However, their use is not that simple.

The use of ANNs will depend on the following aspects:

The data representation and its application will determine the choice of model.
The selection of the learning algorithm.
The resulting ANN can be extremely robust, by selecting the appropriate model.

Applications of ANN:

Systems identification and control. Examples: vehicle control, trajectory prediction, process control, natural resource management.
Quantum chemistry, games, and decision-making. Examples: chess, poker, etc.
Pattern recognition. Examples: radar systems, facial recognition, signal classification, sequence recognition, among others.
Handwritten text recognition, medical diagnosis. For example, automated systems for commerce in various sectors of activity, data mining, etc.
Visualization, machine translation, differentiating between desired and unwanted reports on social media. Examples: spam prevention, etc.
Diagnosis of various types of cancer. Examples: lung cancer, prostate cancer, and colorectal cancer. Thus, these networks could predict multiple outcomes for patients from related institutions, among other things.

ANN Classification:

The single-layer neural network is the simplest network.
The generalization of the single-layer neural network is considered multilayer.
Convolutional with the multilayer perceptron, it only joins a subset of them.
They do not have a layered structure and only allow arbitrary connections between neurons; they are called recurrent.
Radial basis networks calculate the function’s output based on the distance to a point called the center.

How to train neural networks?

Training involves instructing a neural network to perform a task. First, neural networks learn by processing large amounts of labeled or unlabeled data.
Second, machine learning software finds patterns in the existing data and uses them to interpret new data to make intelligent decisions.

Therefore, ANNs are important because they can help computers make intelligent decisions with limited human assistance. Additionally, it can be used naturally in online learning and large dataset applications, such as diagnosing various types of cancer.