Download presentation
Presentation is loading. Please wait.
1
1-norm Support Vector Machines Good for Feature Selection Solve the quadratic program for some : min s. t.,, denotes where or membership. Equivalent to solve a Linear Program as follows:
2
-Support Vector Regression (Linear Case:) Given the training set: Motivated by SVM: should be as small as possible Some tiny error should be discarded Represented by an matrix and a vector Try to find such that that is where
3
-Insensitive Loss Function -insensitive loss function: The loss made by the estimation function, at the data point is If then is defined as: (Tiny Error Should Be Discarded)
4
-Insensitive Linear Regression Find with the smallest overall error
5
Five Popular Loss Functions
6
-Insensitive Loss Regression Linear -insensitive loss function: where is a real function Quadratic -insensitive loss function:
7
- insensitive Support Vector Regression Model Motivated by SVM: should be as small as possible Some tiny error should be discarded where
8
Why minimize ? probably approximately correct (pac) Consider performing linear regression for any training data distribution and then Occam’s razor : the simplest is the best
9
Reformulated - SVR as a Constrained Minimization Problem subject to n+1+2m variables and 2m constrains minimization problem Enlarge the problem size and computational complexity for solving the problem
10
SV Regression by Minimizing Quadratic -Insensitive Loss We have the following problem: where
11
Primal Formulation of SVR for Quadratic -Insensitive Loss Extremely important: At the solution subject to
12
Dual Formulation of SVR for Quadratic -Insensitive Loss subject to
13
KKT Complementarity Conditions KKT conditions are : Don ’ t forget we have:
14
Simplify Dual Formulation of SVR subject to The case, problem becomes to the least squares linear regression with a weight decay factor
15
Kernel in Dual Formulation for SVR Then the regression function is defined by Supposesolves the QP problem: where is chosen such that with subject to
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.