This page describes the third of six project milestones.
You are encouraged to discuss these items with the professor and other students as you are working on this assignment. I am glad to help you with ideas on any parts that you’re finding especially challenging.
This should be one of the last things you write. For an assignment of this scope, it should be less than a page in length, highlighting what was done and summarizing key results.
State the goal of the project including any key metrics (e.g., desired accuracy, number of classes).
If you have a network training run, include a summary of that, even if the results are not good and there are problems with the data. In most cases, you’ll want to include a graph of loss and/or error. If you are doing a classification project, you will usually want to include a confusion matrix. For other types of projects, include some discussion and illustration of example results.
If you are earlier in the process, such as organizing/understanding/preprocessing data, include an analogous level of detail showing the reader your progress and challenges with this work.
Include a clear and complete description of your network structure (if it is complex, you may summarize key parts, especially if they come from standard structures like VGG19 or RESNET). In addition or alternatively (whichever you think will be clearer for the reader), include a diagram of your network from your design tool or the text equivalent (e.g., Keras model summary text).
Update this from the previous milestone, answering the same general questions. Highlight and discuss any changes that might surprise the reader, e.g., if you needed to go in a different direction with your data source or decided to use a different augmentation technique.
You have roughly 4 weeks remaining to complete the project, including preparing a presentation of your work in progress and writing up your final results. Put together a plan for the remaining time. For each week, highlight about 2-3 goals that you expect to complete (e.g., load additional data, try particular variations to network structure depending on early results, compare 2 training methods, prepare ROC or precision-recall curve for final classification system).