| Title |
A conjugate gradient method for two dimensional scaling / |
| Authors |
Žilinskas, Antanas ; Jakaitienė, Audronė |
| Full Text |
|
| Is Part of |
Communication and cognition.. Gent. 2010, vol. 43, no. 1-2, p. 3-13.. ISSN 0378-0880 |
| Keywords [eng] |
Visualization ; optimization ; multidimensional scaling ; local descent ; conjugate gradient method ; convergence |
| Abstract [eng] |
Multidimensional scaling is a technique for representing multidimensional data (a set of points in a multidimensional space) in a space of lesser dimensionality. The case of the 2–dimensional embedding space is of a special interest since 2-dimensional images are well suitable for visualization. The quality of visualization is measured by the difference between the pair wise distances in the original and embedding spaces defined by the STRESS function. The latter should be minimized. This complicated (multimodal) minimization problem can be tackled by a hybrid method combining a genetic type algorithm with a conjugate gradient descent routine. In the present paper a version of conjugate gradient method oriented to STRESS minimization is considered. |
| Published |
Gent |
| Type |
Journal article |
| Language |
English |
| Publication date |
2010 |
