1: \begin{abstract}

2: This paper deals with unsupervised clustering with feature selection in high dimensional space. The problem is to estimate both labels and a sparse projection

3: matrix of weights. To address this combinatorial

4: non-convex problem maintaining a strict control on the sparsity of the matrix of

5: weights, we propose an alternating minimization of the Frobenius norm criterion.

6: We provide a new efficient algorithm named k-sparse which alternates k-means

7: with projection-gradient minimization. The projection-gradient step is a method of

8: splitting type, with exact

9: projection on the $\ell^1$ ball to promote sparsity.

10: The convergence of the  gradient-projection step is addressed,

11: %Each iterative step of our algorithm  necessarily lowers the

12: %cost  and the the cost monotonically decrease. 

13: and a preliminary analysis of the alternating minimization is made.

14: %The Frobenius norm criterion converges as the number of

15: %iterates in Algorithm k-sparse goes to infinity.

16: Experiments on Single Cell RNA sequencing datasets show that our method significantly improves the

17: results of PCA k-means, spectral clustering, SIMLR,

18: and Sparcl methods. The complexity of our method is linear in the number of

19: samples (cells), so that the method scales up to large datasets.

20: \end{abstract}

21: