1: \begin{abstract}

2: We study the problem of optimal power allocation in single-hop multi-antenna ad-hoc wireless networks. 

3: A standard technique to solve this problem involves optimizing a tri-convex function under power constraints using a block-coordinate-descent (BCD) based iterative algorithm. This approach, termed WMMSE, tends to be computationally complex and time consuming. 

4: Several learning-based approaches have been proposed to speed up the power allocation process. 

5: A recent work, UWMMSE, learns an affine transformation of a WMMSE parameter in an unfolded structure to accelerate convergence. 

6: In spite of achieving promising results, its application is limited to single-antenna wireless networks. 

7: In this work, we present a UWMMSE framework for power allocation in (multiple-input multiple-output) MIMO interference networks.

8: Through an empirical study, we illustrate the superiority of our approach in comparison to WMMSE and also analyze its robustness to changes in channel conditions and network size.          

9: \end{abstract}

10: