Final Report Peer Review

[camilocedeno]

In your project, you test the limitations of deep learning on a simple combinatorial optimization problem, the Minimum Spanning Tree problem. There exists simple greedy algorithms that are guaranteed to get you the optimal solutions, however, you are interested in determining how effective deep learning models are at solving the MST problem. You train a GNN by using Google's DeepMind OpenSpiel framework. To train and test the model. You generate synthetic graphs and find the optimal solution using a known greedy algorithm. You test your model's performance by determining if it produces a Spanning Tree and if it does, how far from the MST it is.

Additional Comments:
I think your team had a very interesting problem. I often hear how deep learning models are "black box" models that can solve lots of complex problems. Therefore, it was interesting to see your results in how well your model could solve a simple problem where the optimal solution can easily be found.
I do not have a large understanding of Neural Networks but I found your paper very interesting to read because it brought the content down to a level I could understand.
Your analysis of the testing you conducted was clear and insightful.
I would have liked to see you include a section about how feasible it was for you to test your model. For example, how long it took to train the model on such a large dataset.