This is an extensive introduction, where we are going to know what are deep neural networks, we are going to know we mean when we talk about a feed forward neural network, what are they used for, when we use them either for classification or to make predictions, we are going to see that neural networks can solve non-linear problems. We are going to end up with the best approach for a feed forward neural network which is a neural network with only 2 or 3 hidden layers, that the activation function that best works are a Leaky RELU, that ADAMS is the best method for Stochastic Optimization which is the way that our neural networks learn base on our data.
In this blog, I am going to focus on Feed Forward Neural Network which is used to make classifications and predictions. It is used also in deep nets like convolutional neural networks as the final layer. It`s also important to know the different architectures for neural networks that exit. It is important because depending on the problem you want to solve you need to look for an architecture that is more suitable for solving a certain kind of problem.
Architectures of Deep Neural Networks
In the field of deep neural networks exits differents architectures, each architecture works better for certain kinds of problems. For example, Convolutional Neural Networks is a neural network architecture, it is used for visual classification and object detection.
Feed Fordward Neural Networks
They are nodes grouped in rows and they are connected with each other between each row. Between each connection, there is a weigh that multiply the exit from the node with the weight and there is a bias that is summed.
This is the architecture.
David Mata blog 