Preceeding the development of this program, I took an online course to familiarize myself with TensorFlow. I was comfortable enough with machine learning concepts, and I have a strong background in mathematics to provide a solid foundation.
The metrics that best describe the performance of my program is its accuracy, which I implemented using the reduce mean function. This metric is a solid choice because it punishes predictions with much higher errors. The reduce mean function aims at lowering the model's variance.
Bias is also a another metric to measure, but given the size and scope of the MNIST dataset I chose not to use it.