Weighted Sparse Simplex Representation (WSSR)

In this repository, we provide the MATLAB implementation for Weighted Sparse Simplex Representation.

We provide two implementations for the full WSSR problem (WSSR): a quadratic programming (QP) implementation, and a projected gradient descent implementation (PGD). We also provide an implementation for solving a subproblem of WSSR (WSSR-LE) by setting a system of linear equations.

There are two specifications for each problem:

• WSSR_cos.m, WSSR_PGD_cos.m and WSSR_le_cos.m use absolute cosine similarity to construct the weight matrix in the case of linear subspaces;

• WSSR_euclid.m, WSSR_PGD_euclid.m and WSSR_le_euclid.m use Euclidean distances to construct the weight matrix in the case of affine subspaces and manifolds.