Scribd Logo
Upload

Welcome to Scribd!

  • 27 views

    Principal Component Analysis

    Uploaded by

    Rashedul Islam
    Principal component analysis (PCA) is a technique used to reduce the dimensionality of large datasets by transforming correlated variables into a smaller number of uncorrelated variables called principal components. PCA can be performed by a perceptron using Hebbian learning rules, which update the perceptron's weights based on correlations in the input data. Specifically, Oja's rule causes the perceptron to identify the direction of maximum variance in the input space, which corresponds to the first principal component. Repeated application of this process extracts all the principal components and performs PCA.

    Copyright:

    © All Rights Reserved
    Download as PPT, PDF, TXT or read online from Scribd

    Principal Component Analysis

    Uploaded by

    Rashedul Islam
    27 views25 pages
    Principal component analysis (PCA) is a technique used to reduce the dimensionality of large datasets by transforming correlated variables into a smaller number of uncorrelated variables called principal components. PCA can be performed by a perceptron using Hebbian learning rules, which update the perceptron's weights based on correlations in the input data. Specifically, Oja's rule causes the perceptron to identify the direction of maximum variance in the input space, which corresponds to the first principal component. Repeated application of this process extracts all the principal components and performs PCA.

    Original Title

    Pca 2

    Copyright

    © © All Rights Reserved

    Available Formats

    PPT, PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Principal component analysis (PCA) is a technique used to reduce the dimensionality of large datasets by transforming correlated variables into a smaller number of uncorrelated variables called principal components. PCA can be performed by a perceptron using Hebbian learning rules, which update the perceptron's weights based on correlations in the input data. Specifically, Oja's rule causes the perceptron to identify the direction of maximum variance in the input space, which corresponds to the first principal component. Repeated application of this process extracts all the principal components and performs PCA.

    Copyright:

    © All Rights Reserved
    Download as PPT, PDF, TXT or read online from Scribd
    Download as ppt, pdf, or txt
    27 views25 pages

    Principal Component Analysis

    Uploaded by

    Rashedul Islam
    Principal component analysis (PCA) is a technique used to reduce the dimensionality of large datasets by transforming correlated variables into a smaller number of uncorrelated variables called principal components. PCA can be performed by a perceptron using Hebbian learning rules, which update the perceptron's weights based on correlations in the input data. Specifically, Oja's rule causes the perceptron to identify the direction of maximum variance in the input space, which corresponds to the first principal component. Repeated application of this process extracts all the principal components and performs PCA.

    Copyright:

    © All Rights Reserved
    Download as PPT, PDF, TXT or read online from Scribd
    Download as ppt, pdf, or txt
    of 25

    Principal component analysis

    Hypothesis: Hebbian synaptic


    plasticity enables perceptrons
    to perform principal
    component analysis
    Outline
    Variance and covariance
    Principal components
    Maximizing parallel variance
    Minimizing perpendicular variance
    Neural implementation
    covariance rule
    Principal component
    direction of maximum variance in the
    input space
    principal eigenvector of the covariance
    matrix
    goal: relate these two definitions
    Variance
    A random variable
    fluctuating about
    its mean value.


    Average of the square of the
    fluctuations.

    ox = x x

    ox
    ( )
    2
    = x
    2
    x
    2
    Covariance
    Pair of random
    variables, each
    fluctuating about
    its mean value.



    Average of product of fluctuations.

    ox
    1
    = x
    1
    x
    1
    ox
    2
    = x
    2
    x
    2

    ox
    1
    ox
    2
    = x
    1
    x
    2
    x
    1
    x
    2
    Covariance examples
    Covariance matrix
    N random variables
    NxN symmetric matrix


    Diagonal elements are variances

    C
    ij
    = x
    i
    x
    j
    x
    i
    x
    j
    Principal components
    eigenvectors with k
    largest eigenvalues



    Now you can calculate
    them, but what do they
    mean?
    Handwritten digits

    Principal component in 2d
    One-dimensional projection
    Covariance to variance
    From the covariance, the variance of
    any projection can be calculated.
    Let w be a unit vector

    w
    T
    x
    ( )
    2
    w
    T
    x
    2
    = w
    T
    Cw
    = w
    i
    C
    ij
    w
    j
    ij

    Maximizing parallel variance


    Principal eigenvector of C
    the one with the largest eigenvalue.

    w
    *
    = argmax
    w: w =1
    w
    T
    Cw

    max
    C
    ( )
    = max
    w: w =1
    w
    T
    Cw
    = w
    *T
    Cw
    *
    Orthogonal decomposition

    Total variance is conserved
    Maximizing parallel variance =
    Minimizing perpendicular variance
    Rubber band computer

    Aw= qyx
    Aw= q y y
    ( )
    x x
    ( )
    Correlation/covariance rule
    presynaptic activity x
    postsynaptic activity y
    Hebbian
    Stochastic gradient ascent
    Assume data has zero mean
    Linear perceptron

    y = w
    T
    x

    Aw= qxx
    T
    w

    E =
    1
    2
    w
    T
    Cw
    C = xx
    T

    Aw = q
    cE
    cw
    Preventing divergence
    Bound constraints
    Ojas rule

    Aw= q yx y
    2
    w
    ( )
    Converges to principal component
    Normalized to unit vector
    Multiple principal components
    Project out principal component
    Find principal component of remaining
    variance
    clustering vs. PCA
    Hebb: output x input
    binary output
    first-order statistics
    linear output
    second-order statistics
    Data vectors
    x
    a
    means ath data vector
    ath column of the matrix X.
    X
    ia
    means matrix element X
    ia
    ith component of x
    a

    x is a generic data vector
    x
    i
    means ith component of x
    Correlation matrix
    Generalization of second moment

    x
    i
    x
    j
    =
    1
    m
    X
    ia
    a=1
    m

    X
    ja

    xx
    T
    =
    1
    m
    XX
    T

    You might also like