Neural Networks

Neural Networks are an essential field in Artificial Intelligence. They are widely used in cutting-edge technology. Our smartphones have a lot of AI parts that helps us to make some tasks in an efficient way. Google translator, Siri, fingerprint and face recognition have neural networks behind.

Artificial Neural Networks are computing systems inspired by the biological neural networks that constitute animal brains

Wikipedia (Artificial Neural Networks)

The first implementation of Neural Network came up in the late. It was a very simple Neural Network that during the years have improved a lot. Now we have different Neural Networks structures and training models depending on the field we want to focus on.

Convolutional Neural Networks, Feed Forward Neural Networks, Recurrent and Neural Networks are widely used in 2019. Each network have a different way to learn and predict. In the next posts I will focus on each of them.

The general idea is that most of the networks have an output layer, some hidden layers and an input layer. The predictions are located in the output layer. Each layer is connected with the other in a particular way.

Independently from the architecture of the network we can train it in a different way.

Supervised Learning

We know what the output must be when we train our network. With the desired output we calculate the overall error and adjust the connections to improve the accuracy of the network. Ideal for regression and classification problems such us image recognition.

In the next post we will see some practical uses and the calculations we must carry out.

Unsupervised learning

We don’t know the solutions and consequently we don’t have labelled data (the inputs and the outputs it must give). It’s used for anomaly detection and clustering.

Reinforcement learning

The idea of this type of learning is to improve the performance in the next generations. The individuals of a population take actions towards an specified goal. If they get the goal or they aproximate to it they get a reward. The overall aim of the network is to give the best rewards to the individuals because it means they are behaving well. The most used algorithm is the Genetic Algorithm that is widely used in the game AI but is not as accurate as the supervised learning.

With this information you have a general idea of what networks are. Now you must put it into practice. I have various tutorials in the webpage such us using Neural network for autonomous driving cars with Genetic Algorithms, digit recognition with Supervised Learning and much more.

Neural Networks can be used in differents sector of reasearch such as medicine, driving and mobiles. In the next video you will see some applications of neural networks with different types of learning:

An introduction of neural networks and its uses nowadays can be seen in this video:

You can also see a deep explanation of feed-forward neural networks or convolutional neural networks