1: \begin{abstract}

2: Sparse Partial Least Squares (sPLS) is a common dimensionality reduction technique for data fusion,

3: which projects data samples from two views by seeking linear combinations with a small number of variables with the maximum variance.

4: However, sPLS extracts the combinations between two data sets with all data samples so that it cannot detect latent subsets of samples.

5: To extend the application of sPLS by identifying a specific subset of samples and remove outliers,

6: we propose an $\ell_\infty/\ell_0$-norm constrained weighted sparse PLS ($\ell_\infty/\ell_0$-wsPLS) method for joint sample and feature selection,

7: where the $\ell_\infty/\ell_0$-norm constrains are used to select a subset of samples.

8: We prove that the $\ell_\infty/\ell_0$-norm constrains have the Kurdyka-\L{ojasiewicz}~property so that a globally convergent algorithm is developed to solve it.

9: Moreover, multi-view data with a same set of samples can be available in various real problems.

10: To this end, we extend the $\ell_\infty/\ell_0$-wsPLS model and propose two multi-view wsPLS models for multi-view data fusion.

11: We develop an efficient iterative algorithm for each multi-view wsPLS model and show its convergence property.

12: As well as numerical and biomedical data experiments demonstrate the efficiency of the proposed methods.

13: \end{abstract}

14: