Please submit your homework in pdf format to the CS420 folder in the following FTP. File name should be like this: 0123456789_tushikui_hw1.pdf.
1 (10 points) PCA algorithm
Give at least two algorithms that could take data set X = fx1; : : : ; xN g, xt 2 Rn 1; 8t as input, and output the first principal component w. Specify the computational details of the algorithms, and discuss the advantages or limitations of the algorithms.
2 (10 points) Factor Analysis (FA)
Calculate the Bayesian posterior p(yjx) of the Factor Analysis model x = Ay + + e, with p(xjy) = G(xjAy + ; e), p(y) = G(yj0; y), where G(zj ; ) denotes Gaussian distribution density with mean and covariance matrix .
(10 points) Independent Component Analysis (ICA)
Explain why maximizing non-Gaussianity could be used as a principle for ICA estimation.
(50 points) Dimensionality Reduction by FA
Consider the following Factor Analysis (FA) model,
x = Ay + + e;
p(xjy) = G(xjAy + ; 2I);
p(y) = G(yj0; I);
where the observed variable x 2 Rn, the latent variable y 2 Rm, and G(zj ; ) denotes Gaussian distribution density with mean and covariance matrix . Write a report on experimental comparisons on model selection performance by BIC, AIC on selecting the number of latent factors, i.e., dim(y) = m.
Specifically, you need to randomly generate datasets based on FA, by varying some setting values, e.g., sample size N, dimensionality n and m, noise level 2, and so on. For example, set N = 100; n = 10; m = 3; 2 = 0:1; = 0, and assign values for A 2 Rn m. The generation process is as follows:
Randomly sample a yt from Gaussian density G(yj0; I), with dim(y) = m = 3;
Randomly sample a noise vector et from Gaussian density G(ej0; 2I), with 2 = 0:1,
et 2 Rn;
Get xt = Ayt + + et.
Collect all the xt as the dataset X = fxtgNt=1.
The two-stage model selection process for BIC, AIC is as follows:
Stage 1: Run EM algorithm on each dataset X for m = 1; :::; M, and calculate the log-likelihood
value ln[p(Xj m)], where m is the maximum likelihood estimate for parameters;
Stage 2: Select the optimal m by
m = arg maxm=1;:::;M J(m);
JAIC (m) = ln[p(Xj k)]
JBIC (m) = ln[p(Xj k)]
You may set M = 5, if you generate the dataset X based on n = 10; m = 3.
The following codes might be useful.
5 (20 points) Spectral clustering
Use experiments to demonstrate that when spectral clustering works well, when it would fail.
Summarize your results.
The following codes might be helpful.