Multiview partial least squares

- We have proposed a multiview learning based partial least squares.
- This model finds a series of direction vectors wi(i=1,..,n), which guarantee covariance between response Y and weighted component ∑i=1NαiXiwi and pairwise correlation of component Xiwi(i=1,…,n reach maximum.
- We have also developed a solution to the proposed method.
- Convergence analysis is conducted in this paper.

In practice, multiple distinct features are need to comprehensively analyze complex samples. In machine learning, data set obtained with a feature extractor is referred as a view. Most of data used in practics are collected with various feature extractors. It is practical to assume that an individual view is unlikely to be sufficient for effective analyzing the property of the sample. Therefore, integration of multiview information is both valuable and necessary. But, traditional partial least squares is proposed for single view high dimensional data modeling,which is invalid for multiview data. In this paper, multiveiw partial least squares is proposed. This model finds a series of direction vectors which guarantee covariance between response and weighted component reach maximum as well as pairwise correlation of component. We then proposed an algorithm for multiview partial least squares. Convergence and bound discussion are also given. Experiments demonstrate that proposed multiview partial least squares is an effective and promising algorithm for practical applications.