Quantum Gaussian Mixture Models - Week 6

This week, I implemented QGMM. In this implementation, the goal is to minimize the objective function (16) in the paper. The mentor recommends me to try the objective function optimization with approximation constant like Lagrange multiplier.

Note

The source code is at https://github.com/KimSangYeon-DGU/GSoC-2019/tree/master/Research/Optimization/Experiments

There are 3 python files and 1 dataset file. In the qgmm_utils.py has some utils for running QGMM and in the draw_utils.py is a utils function to draw the results of training. Lastly, the main.py is to test the performance of training.

The objective equation to minimize is NLL + lambda * sum of probabilities - 1.

To make QGMM more robust, some methods are necessary to initialize parameters properly and solve positive definite covariance problems in Cholesky decomposition.

Thanks for reading :)