Oct 9, 2009

NOTE 0910

1. BDeu prior, 是 dirichlet,不过有某种特性
likelihood equivalent uniform Bayesian Dirichlet

保证 Markov equivalent models should have the same marginal likelihood
fullbnt 提供了这个函数

2. Quaternion
Basic opeartion
norm


3. Quaternion optical flow
1) /|q1||q2|=1, q2 = q1+dq1*(dx,dy,dt) 这个式子没法化简,我使用了matlab symbolic toolbox
syms r g b rx ry rt gx gy gt bx by bt dx dy dt

r2 = r + rx*dx + ry*dy + rt*dt;
g2 = g + gx*dx + gy*dy + gt*dt;
b2 = b + bx*dx + by*dy + bt*dt;

f = (r*r2 + g*g2 + b*b2)^2 - (r^2+g^2+b^2)*(r2^2+g2^2+b2^2);

eqa = expand(f);
最后结果很复杂

2). Quaternion least square

3). Multiple instances learning
Software 1 MILL: A Multiple Instance Learning Library
* Iterated-discrim APR [Dietterich et al., 1997]
* Diverse Density (DD) [Maron and Lozano-Perez, 1998]
* EM-DD [Zhang and Goldman, 2001]
* Two SVM variants for MIL [Andrews et al., 2002]
* Citation-kNN for MIL [Wang and Zucker, 2000]
Software 2 Multiple Instance Learning 
这个比较新(2007),有参考论文,代码基于 spider, 粗粗看过

DATASET
1. 

4) multi-channel SVM based on libsvm
应该没有什么花头
方法一:给定 kernel value, refer to the README file
方法二:修改 k_function 和 kernel_function

5) pyramid match kernel
two papers, one uniform pyramid, the other vocabulary guided pyramid.

0 comments: