Overview of Neural Networks

Overview of Neural Networks

Photo by Alina Grubnyak / Unsplash

Artificial Intelligence is so popular in our world. If you are interested in AI, You would probably know the existence of an Artificial Neural Network. Do you wanna know basically what is a neural network? Then come and dive into NeuralNets by this article.

Table of Contents

What is Neural Network
Which areas do we use NNs?
How does a Neural Network work?
Scale of Neural Networks

What is Neural Network

Neural Network, is just simulated and inspired by real neurons. It is a Supervised System made up of neurons and layers. Every neuron connected with the before layer’s neurons through this Neural Network can learn and solve complex problems. Neural Networks have the ability; Learn and Solve.


Which areas do we use NNs?

Companies such as Visa, Master Cards, American Express, Europen Banks, USA Banks, ASIA Pacific Institution, High Street Banks, and Brokers in the UK have hugely invested in neural network technology. An article in The Economist magazine proved with numerical data that sales and internal volume including neural network technology have been on the rise in recent years.

Neural Networks can be applied in a lot of areas such as Finance, Engineering, Medicine, Manufacture. Neural networks can detect damage possibilities for a product before happened. Also, NN’s Prediction abilities proved in finance areas. Neural Networks can work effectively in areas such as macroeconomic predicting, credit scoring, currency forecasting, and risk analysis.

Finance: Fraud Detection, Credit Scoring
Computer Vision: Face-Motion-Object-Emotion Recognitions
Bioinformatics: Tumor Detection, Cancer Detection, Stroke Detection
Energy Production: Price and Load Forecast
Automotive, Aerospace: Predictive Maintenance
Natural Language Processing: Voice Assistant, Machine Translation, Speech Recognition, Chatbots …

How does a Neural Network work?


Neural Network, is a mathematical system composed of a lot of connected neurons. Artificial neurons are connected with each other, it is like Biological Neurons. Just like them, they receive input signals, collect these signals, process them, and deliver the outputs.

Neural Network has five sections:
Inputs: Its inputs that NN takes first
Weights: Our weight parameter is our NN’s most important parameter
Addition: Inputs and weights are multiplied and we add our bias parameter
Activation Function: We have to know our neuron’s activity
Outputs: Output of the activation func. is our neuron’s output. Every neuron would have a lot of input but the output must have 1 value.

The scale of Neural Networks

Just have two parts Shallow and Deep Neural Networks.

Shallow Neural Networks:

“Shallow” has just 3 layers for neurons.

  • Input layer
  • Hidden layer (Math processes are happening here)
  • Output layer (Our statistical result)

In brief, “shallow” neural networks are generally used for basic problems. Because with just 1 hidden layer you can’t extract specific features.

Some articles express deep neural networks have given better results than shallow neural networks. We can say that the hidden layers are effective ways for feature extraction.

Deep Neural Networks:

It is the developed form of neural networks and its hidden layers are more than one. Ian Goodfellow, Bengio and Courville proved neural networks would solve the complex problem but deep neural networks have better accuracy because when you add more layers you would have better accuracy. Hidden layers are useful until 9–10 layers after your prediction will worse. Nowadays, most neural network models and applications have deep neural networks with 3–10 layers.

Andrej Karpaty
Gradient descent can write code better than you. I’m sorry. -Andrej Karpathy

There are a lot of deep learning architecture. These architectures can integrate a lot of areas and have good results.

CNN(Convolutional Neural Network), RNN(Recurrent Neural Network), GAN(Generative Adversarial Network), LSTM(Long-Short-Term-Memory) is some type of advanced level architectures. GoogleNet, VGG-16, AlexNet, LeNet, U-Net some of deep learning models.

About author 🧙‍♂ ️

I like communicating with people online and on social media as a self-taught machine learning engineer. Follow me to be my friend, and I look forward to engaging with you here or on other platforms! Long details of who I am may be found here!🔥