CSC4651/5651 Deep Learning In Signal Processing: Project Milestone 6: Final Results

This page describes the final milestone out of six.

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.

Overview

You will write a research paper covering your project. It will have the following standard sections, in order:

Some of these sections will likely be updated and expanded versions of items you’ve previously written, especially for your background paper and preliminary results.

If you have additional details that do not fit neatly in the above format you may either insert sections if the material is essential to the flow of your explanation, or you may add appendices at the end if the information is not essential to the flow but may be of interest to some readers.

Abstract

The abstract is a brief summary of the entire paper. It is used by readers to decide if they want to read further. It is typically a single paragraph and must at least briefly state the problem, how it was solved, and summarize the results (e.g., state accuracy and precision).

Introduction

The introduction serves two purposes. First, it introduces the problem that you solved and describes it (some details of the description may be deferred to the background section if they fit there and are not essential to a general understanding). Second, it briefly (usually in 1 paragraph) summarizes the the remaining sections of the paper giving the reader a high level view of what they will find in each section, e.g., “In the results section we show that we were able to achieve a training accuracy of 97% with a precision of 90%.”

Background

This section will likely borrow heavily from your background paper. Here you summarize and cite the existing approaches to solving the problem on which you’re building. Also, you summarize and cite any key references that enable the work to be done such as special loss metrics or references of special use in your particular domain (e.g., human hearing, steganography).

Here and throughout the paper, you should use in-line citations (also called in-text citations), that is, brief citations right in the text next to the relevant idea that can be used to look up the full citation in your references section. You may use any standard format you wish (APA, MLA, etc.) but may want to consider IEEE style, which is common in this subject area.

Methods

Here you discuss the various approaches you used to solve the problem, perhaps including a few variations that you tried to improve performance. Diagrams or other illustrations of your network architecture are appropriate here.

Results

In this section, you give quantitative results including any precision-recall curves, confusion matrices, best validation loss, etc. If you tried variations to solving the problem, include them for all or at least key variations.

Discussion

Here you discuss your results and evaluate how successful you were. Are the results useable? Why or why not? If you tried different methods, which performed better, would you have predicted that, and why?

Conclusions

Here you succinctly evaluate the success or lack thereof of your work and give key lessons learned. It is also common to discuss future work or what you think the next steps should be in this area and why.

Appendix: Source Code

Include either or both of a full listing of all your source code and a link to an online version of your code (e.g., git repository).