CS558 Project Local SVM Classification based on triangulation (on the plane) Glenn Fung
Outline of Talk Classification problem on the plane All of the recommended stages were applied: Sampling Ordering: Clustering Triangulation Interpolation (Classification) SVM: Support vector Machines Optimization: Number of training points increased Evaluation: Checkerboard dataset Spiral dataset
Classification Problem in Given m points in 2 dimensional space Represented by an m-by-2 matrix A Membership of each in class +1 or –1
SAMPLING: 1000 randomly sampled points
ORDERING: Clustering A Fuzzy-logic based clustering algorithm was used. 32 cluster centers were obtained
ORDERING: Delaunay Triangulation Algorithms to triangulate and to get the Delaunay triangulation from HWKs 3 and 4 were used. Given a point,the random point approach is used to localize the triangle that contains it.
Interpolation: SVM SVM : Support Vector Machine Classifiers A different nonlinear Classifier is used for each triangle The triangle structure is efficiently used for both training and testing phases and for defining a “simple” and fast nonlinear classifier.
What is a Support Vector Machine? An optimally defined surface Typically nonlinear in the input space Linear in a higher dimensional space Implicitly defined by a kernel function
What are Support Vector Machines Used For? Classification Regression & Data Fitting Supervised & Unsupervised Learning (Will concentrate on classification)
Support Vector Machines Maximizing the Margin between Bounding Planes A+ A-
The Nonlinear Classifier The nonlinear classifier: Where K is a nonlinear kernel, e.g.: Gaussian (Radial Basis) Kernel : The -entry of represents the “similarity” of data pointsand
Reduced Support Vector Machine Algorithm Nonlinear Separating Surface: (i) Choose a random subset matrix of entire data matrix (ii) Solve the following problem by the Newton method with corresponding : min (iii) The separating surface is defined by the optimal solution in step (ii):
How to Choose in RSVM? is a representative sample of the entire dataset Need not be a subset of A good selection of may generate a classifier using very small Possible ways to choose : Choose random rows from the entire dataset Choose such that the distance between its rows exceeds a certain tolerance Use k cluster centers of as and
Obtained Bizarre “Checkerboard”
Optimization: More sampled points Training parameters adjusted
Result: Improved Checkerboard
Nonlinear PSVM: Spiral Dataset 94 Red Dots & 94 White Dots
Next:Bascom Hill
Some Questions Would it work for B&W pictures (regression instead of classification? Aplications?