# Neural Net (AI Studio Core)

## Synopsis

This operator learns a model by means of a feed-forward neural network trained by a back propagation algorithm (multi-layer perceptron). This operator cannot handle polynominal attributes.## Description

This operator learns a model by means of a feed-forward neural network trained by a back propagation algorithm (multi-layer perceptron). The coming paragraphs explain the basic ideas about neural networks, need-forward neural networks, back-propagation and multi-layer perceptron.

An artificial neural network (ANN), usually called neural network (NN), is a mathematical model or computational model that is inspired by the structure and functional aspects of biological neural networks. A neural network consists of an interconnected group of artificial neurons, and it processes information using a connectionist approach to computation (the central connectionist principle is that mental phenomena can be described by interconnected networks of simple and often uniform units). In most cases an ANN is an adaptive system that changes its structure based on external or internal information that flows through the network during the learning phase. Modern neural networks are usually used to model complex relationships between inputs and outputs or to find patterns in data.

A feed-forward neural network is an artificial neural network where connections between the units do not form a directed cycle. In this network, the information moves in only one direction, forward, from the input nodes, through the hidden nodes (if any) to the output nodes. There are no cycles or loops in the network.

Back propagation algorithm is a supervised learning method which can be divided into two phases: propagation and weight update. The two phases are repeated until the performance of the network is good enough. In back propagation algorithms, the output values are compared with the correct answer to compute the value of some predefined error-function. By various techniques, the error is then fed back through the network. Using this information, the algorithm adjusts the weights of each connection in order to reduce the value of the error function by some small amount. After repeating this process for a sufficiently large number of training cycles, the network will usually converge to some state where the error of the calculations is small. In this case, one would say that the network has learned a certain target function.

A multilayer perceptron (MLP) is a feed-forward artificial neural network model that maps sets of input data onto a set of appropriate output. An MLP consists of multiple layers of nodes in a directed graph, with each layer fully connected to the next one. Except for the input nodes, each node is a neuron (or processing element) with a nonlinear activation function. MLP utilizes back propagation for training the network. This class of networks consists of multiple layers of computational units, usually interconnected in a feed-forward way. In many applications the units of these networks apply a sigmoid function as an activation function.

In this operator usual sigmoid function is used as the activation function. Therefore, the values ranges of the attributes should be scaled to -1 and +1. This can be done through the *normalize* parameter. The type of the output node is sigmoid if the learning data describes a classification task and linear if the learning data describes a numerical regression task.

## Input

- training set (Data table)
The input port expects an ExampleSet. It is output of the Retrieve operator in our example process. The output of other operators can also be used as input.

## Output

- model (Improved Neural Net)
The Neural Net model is delivered from this output port. This model can now be applied on unseen data sets for prediction of the

*label*attribute. - example set (Data table)
The ExampleSet that was given as input is passed without changing to the output through this port. This is usually used to reuse the same ExampleSet in further operators or to view the ExampleSet in the Results Workspace.

## Parameters

- hidden layersThis parameter describes the name and the size of all hidden layers. The user can define the structure of the neural network with this parameter. Each list entry describes a new hidden layer. Each entry requires name and size of the hidden layer. The layer name can be chosen arbitrarily. It is only used for displaying the model. Note that the actual number of nodes will be one more than the value specified as hidden layer size because an additional constant node will be added to each layer. This node will not be connected to the preceding layer. If the hidden layer size value is set to -1 the layer size would be calculated from the number of attributes of the input example set. In this case, the layer size will be set to
*(number of attributes + number of classes) / 2 + 1*. If the user does not specify any hidden layers, a default hidden layer with sigmoid type and size equal to*(number of attributes + number of classes) / 2 + 1*will be created and added to the net. If only a single layer without nodes is specified, the input nodes are directly connected to the output nodes and no hidden layer will be used. - training cyclesThis parameter specifies the number of training cycles used for the neural network training. In back-propagation the output values are compared with the correct answer to compute the value of some predefined error-function. The error is then fed back through the network. Using this information, the algorithm adjusts the weights of each connection in order to reduce the value of the error function by some small amount. This process is repeated
*n*number of times.*n*can be specified using this parameter. - learning rateThis parameter determines how much we change the weights at each step. It should not be 0.
- momentumThe momentum simply adds a fraction of the previous weight update to the current one. This prevents local maxima and smoothes optimization directions.
- decayThis is an expert parameter. It indicates if the learning rate should be decreased during learning.
- shuffleThis is an expert parameter. It indicates if the input data should be shuffled before learning. Although it increases memory usage but it is recommended if data is sorted before.
- normalizeThis is an expert parameter. The Neural Net operator uses an usual sigmoid function as the activation function. Therefore, the value range of the attributes should be scaled to -1 and +1. This can be done through the normalize parameter. Normalization is performed before learning. Although it increases runtime but it is necessary in most cases.
- error epsilonThe optimization is stopped if the training error gets below this epsilon value.
- use local random seedIndicates if a
*local random seed*should be used for randomization. - local random seedThis parameter specifies the
*local random seed*. It is only available if the*use local random seed*parameter is set to true.

## Tutorial Processes

### Introduction to Neural Net

The 'Ripley' data set is loaded using the Retrieve operator. A *breakpoint* is inserted here so you can see the data set before the application of the Neural Net operator. You can see that this data set has two regular attributes i.e. att1 and att2. The *label* attribute has two possible values i.e. 1 or 0. Then the Neural Net operator is applied on it. All parameters are used with default values. When you run the process, you can see the neural net in the Results Workspace. There are *x+1* number of nodes in the input, where *x* is the number of attributes in the input ExampleSet (other than *label* attribute). The last node is the threshold node. There are *y* number of nodes in the output, where *y* is the number of classes in the input ExampleSet (i.e. number of possible values of *label* attribute). As no value was specified in the *hidden layers* parameter, the default value is used. Therefore, the number of nodes are created in hidden layer are = size of hidden layer = *(number of attributes + number of classes) / 2 + 1* = *(2+2)/2+1*= *3*. The last node (4th node) is a threshold node.
The connections between nodes are colored darker if the connection weight is high. You can click on a node in this visualization in order to see the actual weights.

This simple process just provides basic working of this operator. In real scenarios all parameters should be carefully chosen.