Viscomp lab FACADE Home: Fast and Automatic Clustering Approach to Data Engineering

Demo accepted by ACM SIGMOD2004



Cluster analysis is a common approach to discovering implicit patterns in huge databases. While many clustering techniques have been developed, it is still challenging to discover natural clusters precisely and efficiently when the data set contains heavy noise and outliers.

FACADE (Fast and Automatic Clustering Approach to Data Engineering) is a powerful tool that can discover clusters of different sizes, shapes, and densities in noisy spatial data. Compared with the existing clustering methods, FACADE has several advantages: first, FACADE produces cleaner results. FACADE uses a more effective noise recognizing and handling method. Both ambiguous and explicit noise can be detected and separated from true data, based on user specification or data density. Second, FACADE requires less parameters and they are easy to set. FACADE provides two modes: expert mode and automatic mode. In expert mode, users can control every step of clustering process interactively through both 2D and 3D visualization support while in automatic mode a learning-based approach is used to fix the algorithm parameters and the clustering process is completely automatic. Third, FACADE is fast. A novel compression method is applied to accelerate the hiearchical merging process. The compression will not lose the patterns but increase the cohesiveness of clusters with a much smaller number of points. As a result, the clustering process can be completed in O(nlogn) time for n data points. FACADE is user-friendly and provides various kinds of visualizations of clustering process and results.

FACADE has recently been extended to accomodate high-dimensional data sets. The extension is needed only in the first step, i.e., the graph construction step. All other steps remain unchanged. The extended FACADE has been applied to cluster microarray and protein structure data. The experiments on benchmarks have got very encouraging results.

FACADE Demo

Please click here to see the demonstration of FACADE on 8 spatial data sets and several screen dumps

FACADE Download

Please click here to download executable programs and movies of FACADE

About Noise Removal (CLEAN)

Please click here to understand CLEAN (CLustering by Eliminating Ambiguous Noise) and how it can remove noise effectively

Secret of Being Fast

Please click here to understand why FACADE can accomplish the hierarchical combination in linear time while traditional combination requires O(n*n).

3-D Final Fantasy

Please click here to see the final clustering results visualized by 3-D graphs produced by FACADE.

Watch Movies

Please click here to see several mpeg4-v2 movies about FACADE (for windows machine and Internet Explorer only).

A Small Flash

Thank you, Gang, for doing this. Flash Plug-in is required to be installed at your machine to see it.

Real-world Applications (Password Required)

We have applied FACADE to some real-world applications like GIS image processing, medical image processing, and microarray gene data processing. Please click here to see them. However, due to copyright issue, password is required here and you can get the password by email request.

Provide Your Data to Us!!!

In FACADE Demo page you can upload your data and run our program online, but this way limits your file size, running time, and data dimensions, and you cannot get customized result as you may expect. Since we cannot provide source code of FACADE to every user who wants it, this page allows you to give your data to us together with your requirements and WE RUN THE PROGRAM FOR YOU and give you the result that meets your possible specific needs.
Basically, we expect any of three kinds of data:
1. Vector data, which could be 2-D or N-D. We accept categorical data only if they can be quantified.
2. Image data, which could be gray level or colorful, gif, jpeg, bmp, png, and pgm files are all acceptable.
3. Graph data, which could be represented in adjacency list/matrix, or GML (graph markup language).
Before you submit your data, you need to register here and our system will automatically assign you an account id. After you have registered, click here to login with your account id.


If you have any question, feel free to contact qianyu AT student.utdallas.edu (Please replace AT with @)


Back to Viscomp Lab