W. Duch, H6429 assignment

Wlodzislaw Duch
H6429 fall 2006 course, first assignment.

We have finished lectures on visualization, so it a good time for your first assignment, due on 1st of October, end of the recess week.

Pick up your favorite mutlivariate visualization method; either one that I have presented or others that you find in the WWW links. Note that the emphasis here is on learning from visualization, not dimensionality reduction or super learning system for regression!
Here you will find some links to visualization methods, more visualization links, links to SOM and its applications. You may also find some new methods browsing the Internet on your own. Search for non-linear dimensionality reduction, isometric mapping; multidimensional scaling; stochastic proximity embedding; reducing the dimensionality of data with neural networks etc.
Many interesting papers with novel methods are below:
manifold learning; more manifold learning;
visualization of hidden nodes in neural networks;
visualization of neural network decisions;
SVM - kernel visualization.
To avoid duplication of the same methods applied to similar problems please let me know what you try to do; send me an email with your name, software name, method and data, and I shall put it on this page and let the others know that it is taken. The assignment should not be a part of your thesis, as this will be an unfair advantage over other students.
Describe the method in details in your formal report or on a Web page; description should be sufficiently detailed to explain the algorithm, understand what the software is doing and how to interpert it, but should not be a verbatim copy from other people's work! The point is to understand the method and then write about it yourself.
Find software that implements it and describe it briefly. If you choose a larger software package, like XGobi, select one or two methods, don't describe the whole package!
Find some interesting multidimensional data: either generate some data yourself, or grab some data from the network. Iris is not an interesting data and you will not learn much from it! Some interesting data sets for visualization are found here; please let me know if you find more interesting data repositories, I shall add it there. You may also have your own data.
Prepare your data (change of format or standardization may be needed), and visualize it.
Analyse the usefulness of the results; what can we learn from this excercise? Perhaps some hypothesis can be proved/disproved, perhaps interesting structure in data will be noticed.

Electronic copy submission is sufficient: please zip or compress all files, give the file your name, write in the email Your-Name, Paper Title, Method Used, Software Used, Data Used (this will be placed in the Table below), and send it to me. Please keep this format to make it easy for me, renaming all your files from "assignment2" etc. to your names is not fun.
If more than one file is send please zip or compress them, give the file Your Name and send it to me.
Try to minimize the size of the file, I want to attach them to this page and I do not have much space in WWW!

Q & A:

1. Is a formal report required? If so, how long should it be?
Your paper is a report. The length should be sufficient for others that have taken this course to be able to understand the method and interpret the results.

2. As visualization is clearer on screen, should we design a webpage instead of writing a report?
Web pages are OK, we may put them into the e-dventure space, but please send me the files.

3. What must we do to score high marks for this assignment? Do we have to study the visualization technique in much greater depth than what was covered in the lectures?
Find an interesting visualization method, a new variant of one of the methods I talked about, or maybe quite new; find interesting data, visualize it, provide interpretation of the visualization, describe what have you learned by doing visualization, and mention potential applications of this knowledge.

Please note that if you did not send me complete data I will not put it into the table!

Topics taken for the first assignment in 2006. Max. number of points is 10

No	Name	Paper	Method	Software	Data	Remarks	Mark
41	Annamala Sarayu Parimal	MDS for ZOO+fMRI data	MDS	XLSTAT	ZOO/FMRI	Very good paper but too much copied verbatim from the Internet	8
30	Ardian Kristanto Poernomo	HiT-MDS for cDNA	HitMDS	HitMDS code+GGobi	Gene-Drug Correlation	Interesting method, but not clear what is the data, what is in figures, what has been learned ...	4
28	Bramandia Ramadhana	Polygonal Line Principal Curves	Principal Curves, Polygonal Line	Kegl Java program	Kegl examples	Descirbes the topic quite well, but misses the point - no intersting data, nothing learned about hte data!	6
39	Chen Tze Chiang	PCA for Boston housing data	PCA	XLStat 2006	Boston housing data	Many errors: covariance formula is wrong! Cov may be negative! It is nonnegative definite matrix. Feature vector is not a matrix. Good analysis of results.	6
1	Chia Yong Sang, Alex	Visualization of High-D Complex Dataset using Relational Perspective Map	Relational Perspective Map	Visumap	Credit scoring (Munich), SFB386	Cox did not developed MDS, otherwise geat theory, data description and analysis.	10
5	Cheu Eng Yeow	Radviz visualization of wine data	Radviz visualization	Orange	Wine	Great paper!	10
21	Chong Yee Seng	Pima-Indian-Diabetes Using PCA/Kernel PCA	PCA	Matlab	Heart Disease, UCI	Kernel PCA is very briefly described; RBC=>RBF; there are many RBF functions, what is RBF kernel? Argument? Overall good data analysis.	7
8	Dang Xuan Hong	PCA in data visualization	PCA	Matlab	Wine + Wisconsin Breast Cancer	Good although general into on PCA; short decription of software; GhostMiner plots are shown but not referenced; 1D PCA looks much better in N-dots plot; which components are most important? What have we learned?	6
11	Han Shuguo	Data Visualization Using FDA/Kernel FDA	FDA and Kernel FDA	Matlab+STPRTool	Ionosphere	How is separab calculated? What are the straigt lines on figures? to separate classes in Fig. 4 line should be to shifted to the right. What can we learn from these figures?	7
37	Hu Meiqun	PCA/MDS for image segmentation	PCA	GhostMiner	Image Segmentation	Very good description of theory & data, interesting experiments	9
27	Hu Meishan	SOM visualization on Glass Identification	SOM	Yale	Glass	Some editorial corrections needed but overall nice paper, well focused on data analysis	8
7	Huang Dong	Wine recognition with FDA	FDA	STPRtool+WD FDA implementation	Wine	Theory is fine, many pictures have been generated, FDA in 2D is suitable for data separation but we have not learned much about the data.	7
20	Huang Yi	Restoring Ink Bleed-Through Degraded Documents	PCA + k-means clustering	Matlab	Own images	Interesting application, well focused	8
10	Iti Chaturvedi	Visualization of Gene Regulatory Networks using Bayesian Networks	Dynamic Bayesian Networks	GeneNetworks	S. Cerevisiae Microarray data (cell-cycle)	Novel for this course, dynamic networks, well done although less focused on data	9
12	John Felix Charles Joseph	Comparison of classical MDS, LTSA & ISOMAP	MDS, LTSA and ISOMAP	Manifold Learning Toolbox, Matlab	Non-Symbolic Features of KDD Cup 1999	Interesting methods; classical scaling is not MDS, interesting data, outliers hsould be removed in Fig.4	8
43	Koh Chin Wei, Eugene	Visualizing Low-level Audio Features Using SOM	SOM	SOM Toolbox 2.0, SDH Toolbox	Audio, own collection	Interesting experiments and very good description of data and methods and great analysis	10
34	Le Minh Nghia	Handwritten digits visualization using Diffusion Map	Diffusion map	MANIfold learning	Handwritten digits	Error in conjugate formula, but intersting method, data and interepretation	9
25	Maggy Anastasia Suryanto	Visualization with Locally Linear Embedding	LLE	Manifold Toolbox	Wisconsin Cancer	Good description of LLE, weaker on interpretation	7
31	Mohamad Hirwan	HitMDS for cDNA on Barley data	MDS	HitMDS	Barley seed expression	No references to methods/data, some symbols not explained, a single experiment made	6
22	Nah Hock Choon, Edwin	Principal curves for hand-written characters	Principal curves	Kegl et al	NIST database of handwritten characters	Good algorithm description, but not much data analysis and learning from data.	6
35	Nai Hong Hwa Francis	Visualization of high-dimensional data with relational perspective map	Relational Perspective Map (RPM)	VisuMap	Ecoli Proteins	traveling abroad ...	0
2	Nguwi Yok Yen	Road Sign Visualization with PCA & Emergent SOM	PCA+Emergent SOM	Databionic ESOM Tools	Roadsigns	Interesting data and well described, although sometimes confusing	9
24	Nguyen Luong Dong	SOM for country data	SOM	MATLAB+SOM Toolbolx	4 continents/4 categories country data	Only basic SOM forms description; some inacurate statments, analysis not too useful	6
15	Nguyen Minh Nhut	Visualization using locally linear embedding	LLE and PCA	Matlab toolbox	Ionosphere	Description of PCA nad LLE up to the point; not much learned about the data itself	7
18	Nguyen Trung Hieu	Learning from visualization	Laplacian Eigenmap	MANIfold learning	Digits with pressure info; Boston housing	Interesting and well described method; not much on software but several nice experiments; ref. 1 incomplete	9
33	Pham Manh Tung	Isometric feature mapping	ISOmap method	MANIfold learning	Pen-Based Handwritten Digits	Rather informal description of methods, several papers quoted; not much on software but not much learned about data	7
23	Phua Si Jie	SOM and ViSOM for Handwritten Digits	Visualization-induced SOM (ViSOM)	SOM+SPRTool Toolbox	Pen-Based Handwritten Digits	Very nice work!	10
3	Puah Wee Choo	Nonmetric Multidimensional Scaling for Visualization	Nonmetric MDS	Matlab EDA toolbox	Shortest inter-depot traveling time	Great description of non-metric MDS; quite unusual data, and interesting analysis	10
32	Ronny	ICA for Blood Vessels Extractions in Retinal Images	ICA	ICALAB	Retinal Images	Fine description of CIA and software; interesting data but not much on experiments and learning about data.	7
19	Sim Sian Hui Kelvin	SOM for stock clustering	SOM	SOM tollbox	Financial ratios of S&P 500 stocks	Good SOM/U-matrix description; many experiments, novel input processing, interesting conclusions.	10
26	Song Hengjie	Visualization with SOM	SOM+scatterograms	Yale	Iris	Methods and tools are fine, but for Iris it is hard to draw interesting conclusions; scatterograms in x3, x4 give more ino ...	7
40	Tan Wi-Meng, Javan	Interpretability of Visualization Techniques Using Scatterograms & Star Glyphs	Scatterograms & Star Glyphs	XmdvTool	Cars	Statistic is wrong: cars have min 3 cylinders not 5.5, no. cylinders and orgin have interesting correlations; good focus on data	6
29	Teng Teck Hou	Statistical Analysis of Forest Cover Type (djvu format)	PCA, KMeans, histograms	Visual C++	Forest Cover Type	Mot much on methods and software, but many figures showing histogram, scatterograms, clusterized scatterograms and PCA, and numerous observations made.	8
17	Tu Tong	Visualizations of Signatures using Laplacian Eigenmaps	Laplacian Eigenmaps	Manifold Learning Toolbox	Signatures, gpdsSIGNATURE database	Interesting method, well described, despite poor feature extraction some lessons are learned.	9
9	Umair Rafique	Visualization of Neural Network Decisions	WD Projection	WD Matlab+Netlab package	Glass and Letters	Interesting experiments to learn about neural network performance rather than data itself.	8
42	Wan Kong Wah	SOM Elucidating Structures in Multimedia Content	SOM	SOM+VOICEBOX Toolboxes	Multimedia WAV data of tennis, classical music and pop song	Short theory but very interesting data and experiments; "fir-elipse? Fur Elise.	10
14	Wang Di	MDS for basketplayers	MDS	Permap	NBA basketball player	Theory described only verbally; detailed software description, intereresting data and experiments	10
16	Wang Lin	Visualization of PCA and FDA on Waveform Data	PCA/FDA	STPRTool	Monk + Waveform	Very good on theory, data well suited for the methods, nice experiments.	10
36	Woo Huizhen, Jessica	Sammon+curvlinear distance for mass spectrometry data	Sammon mapping and Curvilinear data analysis	MZmine	Fatty acid amide hydrolase (FAAH)	Description of CCA and data not too clear (how many peakes has been used?), data interepretation is fine.	7
38	Wu Min	PCA, Kernel PCA and MDS for Zoo data	PCA, Kernel PCA & Classical MDS	STPRTools and XLSTAT	Zoo	Theory kept on basic level, software not described, not much learned from the data	6
4	Yeong Sui Sum	Scatter Plot Matrix	SOM	SOM in Maple or Matlab	Body fat and cirumferences	Not much to say ...	4
6	Zhang Xuejie	PCA/Kernel PCA for Glass identification	PCA & Kernel PCA	STPRTool	Glass data	Great paper in all respects!	10
13	Zhao Guopeng	Analyzing Extreme Learning Machine by Visualization	WD Projection	WD Matlab+Netlab package+ELM Matlab code	K-category data	Interesting problem, great analysis!	10

Wlodzislaw Duch H6429 fall 2006 course, first assignment.

Wlodzislaw Duch
H6429 fall 2006 course, first assignment.