network objects, data, and

Upload: ali-raxa

Post on 06-Apr-2018

226 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/3/2019 Network Objects, Data, and

    1/16

    Network Objects, Data, and

    Training Styles

  • 8/3/2019 Network Objects, Data, and

    2/16

    Introduction

    The work flow for the neural network design process has seven primarysteps:

    1. Collect data

    2. Create the network

    3. Configure the network4. Initialize the weights and biases

    5. Train the network

    6. Validate the network

    7. Use the network

    This chapter shows how to format the data for presentation to thenetwork. It also explains network configuration and the two forms ofnetwork training: incremental training and batch training.

  • 8/3/2019 Network Objects, Data, and

    3/16

    Introduction

    There are four different levels at which the Neural Network Toolboxsoftware can be used.

    The first level is represented by the GUIs and can be launched by using nnstart . These provide a quick way to access the power of thetoolbox for many problems of function fitting, pattern recognition,

    clustering and time series analysis. The second level of toolbox use is through basic command-line

    operations. The command-line functions use simple argument listswith intelligent default settings for function parameters.

    A third level of toolbox use is customization of the toolbox. This

    advanced capability allows you to create your own custom neuralnetworks, while still having access to the full functionality of thetoolbox.

    The fourth level of toolbox usage is the ability to modify any of the M-files contained in the toolbox.

  • 8/3/2019 Network Objects, Data, and

    4/16

    MODEL

    Simple Neuron

    The fundamental building block

    for neural networks is the single-input

    neuron, such as this example.

    There three processes whichtaking place are : the weight function,

    the net input function and the transfer

    function.

  • 8/3/2019 Network Objects, Data, and

    5/16

    Transfer Functions

    Many transfer functions are included in the Neural Network Toolboxsoftware. Two of the most commonly used functions are shown below.

    Linear Transfer Function

    Neurons of this type are used inthe final layer of multilayer networks

    that are used as function approximations.

  • 8/3/2019 Network Objects, Data, and

    6/16

    Neural Network DESIGN One-Input Neuron

    Alter the weight, biasand input by draggingthe triangular shapedindicators.

    Pick the transferfunction w ith theF menu.

    Watch the change tothe neuron functionand its output.

    Chapter 2

    1

    pw

    b

    a

    Input Line ar Ne uron: a = pure lin(w *p+b)

    F:

    -2 0 2

    w

    -2 0 2

    b

    -4 -2 0 2 4-4

    -2

    0

    2

    4

    p

    a

    Neural Network DESIGN One-Input Neuron

    Alter the weight, bias

    and input by dragging

    the triangular shaped

    indicators.

    Pick the transferfunction with the

    F menu.

    Watch the change to

    the neuron functionand its output.

    Chapter 2

    1

    pw

    b

    a

    Input Line ar Ne uron: a = pur elin(w *p+b)

    F:

    -2 0 2

    w

    -2 0 2

    b

    -4 -2 0 2 4-4

    -2

    0

    2

    4

    p

    a

  • 8/3/2019 Network Objects, Data, and

    7/16

    Neuron with Vector Input

    The simple neuron can be extended to handle inputs that are vectors. Aneuron with a singleR-element input vector is shown below. Here theindividual input

    elements are multiplied by weights

    and the weighted values are fed to

    the summing junction.

    This expression can, of course,

    be written in MATLAB code as

  • 8/3/2019 Network Objects, Data, and

    8/16

    Abbreviated Notation

  • 8/3/2019 Network Objects, Data, and

    9/16

    Network Architectures

    One Layer of Neurons

  • 8/3/2019 Network Objects, Data, and

    10/16

    Multiple Layers of Neurons

    To describe networks having multiple layers, the notationmust be extended. Specifically, it needs to make adistinction between weight matrices that are connected toinputs and weight matrices that are connected betweenlayers. It also needs to identify the source and destinationfor the weight matrices.

    We will call weight matrices connected to inputs inputweights; we willcall weight matrices connected to layeroutputs layer weights.

    Further, superscripts are used to identify the source(second index) and the destination (first index) for thevarious weights and other elements of the network.

  • 8/3/2019 Network Objects, Data, and

    11/16

    the weight matrixconnected to the input

    vector p is labeled as aninput weight matrix

    (IW1,1) having a source1 (second index) and adestination 1 (firstindex). Elements of layer

    1, such as its bias, net input,and output have asuperscript 1 to say thatthey are associated with the

    first layer.

  • 8/3/2019 Network Objects, Data, and

    12/16

  • 8/3/2019 Network Objects, Data, and

    13/16

  • 8/3/2019 Network Objects, Data, and

    14/16

    The layers of a multilayer network play different roles. Alayer that produces the network output is called an outputlayer. All other layers are calledhidden layers. The three-layer network shown earlier has one output layer (layer 3)

    and two hidden layers (layer 1 and layer 2). Some authorsrefer to the inputs as a fourth layer. This toolbox does notuse that designation.

  • 8/3/2019 Network Objects, Data, and

    15/16

    Input and Output Processing Functions

    Input Processing FunctionsNetwork inputs might have associated processing functions. Processingfunctions transform user input data to a form that is easier or moreefficient for a network.

    Mapminmax transforms input data so that all values fall into theinterval [1, 1]. This can speed up learning for many networks.

    removeconstantrows removes the rows of the input vector thatcorrespond to input elements that always have the same value, becausethese input elements are not providing any useful information to thenetwork.

    fixunknowns which recodes unknown data (represented in the usersdata with NaN values) into a numerical form for the network. Fixunknowns preserves information about which values are known andwhich are unknown.

  • 8/3/2019 Network Objects, Data, and

    16/16

    Output Processing Functions

    Output processing functions are used to transform user-providedtarget vectors for network use. Then, network outputs are reverse-

    processed using the same functions to produce output data with thesame characteristics as the original user-provided targets.

    Both mapminmax and removeconstantrows are often associatedwith network outputs. However, fixunknowns is not. Unknown values

    in targets (represented by NaN values) do not need to be altered fornetwork use.