Codebase list libocas / HEAD
HEAD

Tree @HEAD (Download .tar.gz) 

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
LIBOCAS - Library implementing Optimized Cutting Plane Algorithm (OCAS) 
solver for training linear SVM classifiers

FEATURES
 - SVM solvers for training linear classifiers from large scale-data.
 - Two-class and genuine multi-class SVM formulations.
 - A stand alone application and MEX interface for Matlab.
 - Optimized for features represented in by different data types: 
   sparse/dense, double/single precision, int8.
 - Reads examples from SVM^light format.
 - Parallelized version of the two-class SVM solver.
 - Allows using different C for each training example (only for Matlab).
 - Training translation invariant image classifier from virtual examples using LBP features.
 - Functions for computing image features based on Local Binary Patterns (LBP).


PROBLEM FORMULATION 

OCAS solver is currently implemented for training binary (two-class) and multi-class
SVM classifiers:

1. Binary case: OCAS solves the following unconstrained convex optimization task
    
   W^*,W0^* = argmin 0.5*(W'*W+W0^2) + sum C(i)*max( 0, 1-y(i)*(W'*X(:,i)+W0*X0) )
                W,W0                  i=1:nData

where C [nData x 1] is the regularization constant for each example, 
X [nDim x nData] are training feature vectors and y [nData x 1] are their
binary labels (+1/-1). The result are parameters W^* [nDim x 1], W0^* [1 x 1] of
the linear rule

      f(X) = sign( X'*W + W0 )

2. Multi-class case: OCAS solves the following unconstrained convex optimization task

   W^*  =  argmin  0.5*sum_y (W(:,y)'*W(:,y)) + C*  sum     max( (y~=y(i)) + (W(:,y) - W(:,y(i)))'*X(:,i)) 
             W                                    i=1:nData  y 

where C is the regularization constant, X [nDim x nData] are training feature
vectors and y [nData x 1] are their labels from 1 to Y. The result are 
parameters W^* [nDim x Y] of the linear rule

      f(X) = argmax X'*W(:,y)
               y

AVAILABILITY

LIBOCAS can be downloaded from 
    http://cmp.felk.cvut.cz/~xfrancv/ocas/html/index.html

PLATFORMS

Developed and tested under GNU/Linux. 

LICENSE

LIBOCAS is licensed under the GPL version 3 (cf. LICENSE).


REFERENCES

V. Franc, S. Sonnenburg. Optimized Cutting Plane Algorithm for Large-Scale Risk Minimization. 
  The Journal of Machine Learning Research (JMLR), vol. 10, pp. 2157--2192. October 2009. 
  http://jmlr.csail.mit.edu/papers/volume10/franc09a/franc09a.pdf

V. Franc, S. Sonnenburg. OCAS optimized cutting plane algorithm for Support Vector 
  Machines. In Proceedings of ICML. Omnipress, 2008.
  http://cmp.felk.cvut.cz/~xfrancv/papers/Franc-OCAS-ICML08.pdf

S. Sonnenburg, V. Franc.  COFFIN: A Computational Framework for Linear SVMs.  
  In Proceedings of the 27nd International Machine Learning Conference (ICML'10). 
  Haifa 2010.
  http://cmp.felk.cvut.cz/~xfrancv/papers/Sonnenburg-COFFIN-ICML10.pdf

Commit History @HEAD