Deep Feed forward Neural Networks

Deep Feed forward Neural Networks

1. Also known as feedforward neural networks, or multilayer perceptrons(MLPs)

   • As it name says, is an a perceptron with multiple layers
   • Are the quintessential deep learning models

2. PERCEPTRON

   • The perceptron is an algorithm for supervised learning of binary classifiers (functions that can decide whether an input, represented by a vector of numbers, belongs to some specific class or not), It is a type of linear classifier,
   • Was invented in 1957 at the Cornell Aeronautical Laboratory by Frank Rosenblatt
   • Was one of the first artificial neural networks to be produced
   • Perceptrons could not be trained to recognise many classes of patterns, due this, the researches invented the DFF NN
   • Can be trained by a simple learning algorithm that is usually called the delta rule

3. Some features

   • Don't form a cycle, are different to recurrent
   • You can build only by combining many layers of single perceptron
   • was the first and simplest type of artificial neural network devised.
   • It can aproximate almost any function
   • The information moves in only one direction, forward, from the input nodes, through the hidden nodes (if any) and to the output nodes.
   • It can resolve non linear problems
   • the overall number of layers gives the DEPTH of the model, the name deep learning arose from this terminology
     • Final layer is the output layer
     • The training examples specify directly what the output layer must do at each point of an input (p)
       • The behavior of the other layers is not directly specified by the training data. this layers are called HIDDEN LAYERS
   • Backpropagation is the most used algorithm to learn in this network

4. Form the basis of many important commercial applications.

   • The convolutional networks used for object recognition from photos are a specialized kind of feedforward network
   • Some examples
     • Airline Marketing Tactician
     • Backgammon
     • Data Compression
     • Driving – ALVINN
     • ECG Noise Filtering
     • Financial Prediction
     • Speech Recognition
     • Sonar Target Recognition

5. Disadvantages

   • Sometimes need a lot of training time
   • it's bad extrapolating
   • The existence of local minimums in the error function makes training difficult