+ All Categories
Home > Documents > L17 My Neural Nw

L17 My Neural Nw

Date post: 06-Apr-2018
Category:
Upload: bomezzz-enterprises
View: 228 times
Download: 0 times
Share this document with a friend

of 32

Transcript
  • 8/3/2019 L17 My Neural Nw

    1/32

    241-320 Design Architecture and Engineeringfor Intelligent System

    Suntorn Witosurapot

    Contact Address:Phone: 074 287369 or

    Email: [email protected]

    January 2011

  • 8/3/2019 L17 My Neural Nw

    2/32

    Lecture 17:

    Introduction to Neural Networks

  • 8/3/2019 L17 My Neural Nw

    3/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 3

    What Are Artificial Neural Networks?

    An extremely simplified model of the brain

    Essentially a function approximator

    Transforms inputs into outputs to the best of its ability

  • 8/3/2019 L17 My Neural Nw

    4/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 4

    What Are Artificial Neural Networks?

    Composed of many neurons that co-operate toperform the desired function

  • 8/3/2019 L17 My Neural Nw

    5/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 5

    What Are They Used For?

    Classification

    Pattern recognition, feature extraction, imagematching

    Noise Reduction

    Recognize patterns in the inputs and producenoiseless outputs

    Prediction Extrapolation based on historical data

  • 8/3/2019 L17 My Neural Nw

    6/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 6

    What Are They Used For?(cont.)

    Examples: Predicting time-series in financial world () Fraud detection (- )

    Object Separation () etc

    Very useful in Data Mining

    imitate a humans ability to learn from experience

    better results are the hope(Adequately designed and trained NN can capture varied patterns)

    Drawback models tend to be difficult to understand(Is it necessary to understand?)

  • 8/3/2019 L17 My Neural Nw

    7/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 7

    What Are They Used For?(cont.)

    is not applying some set formula, but balancing herexperience and knowledge of sale prices of similarhousesher knowledge about housing prices is not

    static...fine tuning her calculation to fit the latest data

    Real EstateAppraiser

  • 8/3/2019 L17 My Neural Nw

    8/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 8

    Loan Prospector ()

    A Neural Network is like a black box that knows how toprocess inputs to create a useful output. The calculation isquite complex and difficult to understand, yet the resultsare often useful

  • 8/3/2019 L17 My Neural Nw

    9/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 9

    Neural Net Attractions

    Ability to learn

    NNs figure out how to perform their function on

    their own

    Determine their function based only upon sampleinputs

    Ability to generalize

    i.e. produce reasonable outputs for inputs it hasnot been taught how to deal with

  • 8/3/2019 L17 My Neural Nw

    10/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 10

    Neural Net Limitations

    Neural Nets are good for prediction and estimationwhen:

    Inputs are well understood

    Output is well understood Adequate examples are trained the neural net

    Neural Nets are only as good as the training set usedto generate it. The resulting model is static and mustbe updated with more recent examples and retrainingfor it to stay relevant

  • 8/3/2019 L17 My Neural Nw

    11/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 11

    Feed-Forward Neural Net Examples

    One-way flow through the network from inputs to outputs

  • 8/3/2019 L17 My Neural Nw

    12/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 12

    Feed-Forward Neural Net Examples(cont.)

  • 8/3/2019 L17 My Neural Nw

    13/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 13

    A Biological neuron

    cell andnucleus

    Axon(Neurit)

    Dendrits

    Synapsis

    Dendrits: (Input)Getting other activations

    Axon: (Output ) forward the activation(from 1mm up to 1m long)

    Synapse: transfer of activation:

    to other cells, e.g. Dendrits of otherneurons

    a cell has about 1.000 to 10.000

    connections to other cells Cell Nucleus: (processing)

    evaluation of activation

  • 8/3/2019 L17 My Neural Nw

    14/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 14

    Natural vs. Artificial Neuron

    cell andnucleus

    Axon(Neurit)

    Dendrits

    Synapsis

    j

    jjii ownet

    ),( iiacti netfact

    )(iouti

    actfo

    w1i w2i wji...

    oi

    net : input from the network

    w : weight of a connectionact : activationfact : activation function :bias/thresholdfout : output function (mostly ID)o : output

  • 8/3/2019 L17 My Neural Nw

    15/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 15

    Abstraction

    Dendrits:weighted (real number) connections

    Axon: output: real number

    Synapse: ---(identity: output is directly forwarded)

    Cell nucleus:unit contains simple functions

    input = (many) real numbersprocessing = activation functionoutput = real number (~activation)

  • 8/3/2019 L17 My Neural Nw

    16/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 16

    Artificial Neuron

  • 8/3/2019 L17 My Neural Nw

    17/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 17

    Loan Prospector()

  • 8/3/2019 L17 My Neural Nw

    18/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 18

    NN can have multiple output neurons

  • 8/3/2019 L17 My Neural Nw

    19/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 19

    How does Neural Net work?

  • 8/3/2019 L17 My Neural Nw

    20/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 20

    Common transfer functions

  • 8/3/2019 L17 My Neural Nw

    21/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 21

    Where Do The Weights Come From?

    The weights in a neural network are the most importantfactor in determining its function

    Training - process of setting the best weights on the

    edges connecting all the units in the network

    Use the training set to calculate weights such that NNoutput is as close as possible to the desired output foras many of the examples in the training set as possible

  • 8/3/2019 L17 My Neural Nw

    22/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 22

    Where Do The Weights Come From?(cont.)

    Network

    changingnetwork parameters

    evaluationnetwork errorlearning

    examples

  • 8/3/2019 L17 My Neural Nw

    23/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 23

    Where Do The Weights Come From? (cont.)

    There are two main types of training

    Supervised Training

    Supplies the neural network with inputs and the desired outputs

    Response of the network to the inputs is measured

    Adjust weights such that differences between desired andactual outputs are minimized

    Unsupervised Training

    Only supplies inputs

    The neural network adjusts its own weights so that similar inputsshould generate the similar outputs

    The network identifies the patterns and differences in theinputs without any external assistance

  • 8/3/2019 L17 My Neural Nw

    24/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 24

    Example: Voice Recognition

    Task: Learn to discriminate between twodifferent voices saying Hello

    Data Sources

    Steve Simpson

    David Raubenheimer

    Format Frequency distribution (60 bins)

    Analogy: cochlea

  • 8/3/2019 L17 My Neural Nw

    25/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 25

    Network architecture

    Feed forward network

    60 input (one for each frequency bin) 6 hidden

    2 output (0-1 for Steve, 1-0 for David)

  • 8/3/2019 L17 My Neural Nw

    26/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 26

    Presentingthe data

    Steve

    David

  • 8/3/2019 L17 My Neural Nw

    27/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 27

    Presenting the data (untrained network)

    Steve

    David

    0.43

    0.26

    0.73

    0.55

  • 8/3/2019 L17 My Neural Nw

    28/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 28

    Calculate error

    Steve

    David

    0.430 = 0.43

    0.261 = 0.74

    0.731 = 0.27

    0.550 = 0.55

  • 8/3/2019 L17 My Neural Nw

    29/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 29

    Backprop error and adjust weights

    Steve

    David

    0.430 = 0.43

    0.261 = 0.74

    0.731 = 0.27

    0.550 = 0.55

    1.17

    0.82

  • 8/3/2019 L17 My Neural Nw

    30/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 30

    Repeat process (sweep) for all training pairs Present data

    Calculate error Backpropagate error Adjust weights

    Repeat process multiple times

  • 8/3/2019 L17 My Neural Nw

    31/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks 31

    Presenting the data (trained network)

    Steve

    David

    0.01

    0.99

    0.99

    0.01

  • 8/3/2019 L17 My Neural Nw

    32/32

    241-320 Design Architecture &Engineering for Intelligent System Introduction to Neural Networks

    32

    Results Voice Recognition

    Performance of trained network

    Discrimination accuracy between knownHellos 100%

    Discrimination accuracy between newHellos

    100%

    Summary:

    Network has learnt to generalise from original data

    Networks with different weight settings can havesame functionality

    Trained networks concentrate on lower frequencies

    Network is robust against non-functioning nodes


Recommended