Final Project Instructions for ECBM E4040

Document Revisions                                                                                                                   1

Final Project Instructions                                                                                                           1

Project Proposals for Topics Outside of the Paper List Provided by the Instructors           5

Document Revisions

●    10/18/2022 - updated for 2022 fall

●    10/14/2021 - updated for 2021 Fall

●   4/9/2021 - updated for 2021 Spring

Final Project Instructions

In a nutshell

A group (team) of up to 3 students needs to: (a) do a comprehensive review of one deep          learning paper, (b) reproduce the results of the paper by recreating the neural network model   and the results using Tensorflow and Python code, (c) document the project and its results in a detailed report, including the comparison between the obtained results and the results in the    original paper.

Option I: Students work on a paper recommended by instructors. The instructors have compiled a list of papers that will provide students with good exposure to applications of deep learning.     See the list of proposed papers in the spreadsheet shared with students. In this case, a project   proposal does not need to be written.

Option II: Students work on a paper chosen by themselves. In this case, a project proposal needs to be written by students, and approved by instructors. See further instructions at the bottom of this document.

Instructions

Teams: Students need to form teams of up to 3 members. Students need to record their project (paper) names and team IDs (project ID/group ID) in the spreadsheet                                          (spreadsheetE4040_20xxFall_studentInformation, tab *.ProjectTeams). If you need to find         partners to complete the team, indicate in the sheet, and work through email and Piazza to find partners.

Tools and Code: Tools that can be used are TensorFlow1.13 (until 2019) or TensorFlow2.0 (or  versions after 2.0 if advised by instructors), and Keras. Code submission has to contain the        requirements file listing the tools used in the project. Tools other than TensorFlow and Keras can not be used. Various papers may have online code, students are not allowed to copy online        code, they must write their own code demonstrating that with many code comments, unique       variable names, etc.

Content of the Report

Final report needs to summarize the original paper, provide a brief review of the relevant         literature, and present the details of the students’ project work and results. Students’ results    need to show figures, plots, tables, running times. In particular, the paper needs to show the   accuracy and differences between the results of the original paper and results obtained by the student group (team). The report needs to provide a comprehensive discussion of the results  and comparison against the original paper: (i) discuss whether you were able to get the same, better or worse results than those described in the paper, (ii) justify why, (iii) provide a              discussion on what can be done to improve your results; (iv) discuss what insights you were   able to gain from your implementation/experimentations. See other details in the grading         summary.

Format of the Report

For the content of the report, the students have to use the report template

“ E4040.20xxFall.ZTCN.report.rt5560.yy4490.zk2172(Template)” (located in google directory       E4040.20xxFall.studentProjects). This is a two-column conference style IEEE paper template    (This template is a modified IEEE Conference Paper Format                                                         https://www.ieee.org/conferences/publishing/templates.html). The first part of the report needs   to describe the original paper - this part should be written in a third person voice (“they” ,             “authors of the original paper” ), whereas the students’ work needs to be written in the first          person voice (“students” , “I” , “We” ).  Throughout the paper, clear statements need to be made    about what E4040 students did, and how their work is similar or different from the original paper. The report should be no less than 4 pages long. In the final section of the report the students     need to summarize, in a table, the contributions of each team member by stating which parts of the work, code and project report have been done by which individual (to show individual           student contributions in fractions).

Report Name

The name of the file containing the report has to follow this format

E4040.20xxFall.teamID.report.UNI1.UNI2.UNI

(example: E4040.2019Fall.ZTCN.report.rt5560.yy4490.zk2172)

●   Project report can be written as a google doc, MS word, latex or other. It must meet the template and content requirements shared with students.

●   Project report needs to be saved in adobe *.pdf format, using the naming format   described above“E4040.2019Fall.teamID.report.UNI1.UNI2.UNI.pdf” . It must be (i) uploaded into the courseworks (every team member needs to upload into her/his  courseworks); (ii) uploaded into the project github as well.

●   Code (which resides in python files and jupyter notebooks) needs to be

uploaded/committed into the private group (team) Github Classroom repo. Follow the instructions on the github classroom - README file. The instructors will enable the    groups (teams) repos, which will have to be named per instructions in the github. The code needs to be runnable by TAs.

●   Submission of the model needs to be committed to the github classroom repo, (for one group (team) there is one github repo)

○    Under the private GROUP (team) github classroom repository for your project

■   The instructors will enable the groups (team) repos, which will have to be named per instructions in the github

●   Datasets: Some datasets will be very large and therefore the datasets should not be       uploaded into the courseworks (this will be penalized). They may also be too big for the  github repo. When datasets are public and available for download over the web, it is       enough to have proper code in python/jupyter notebooks for fetching the code. If the       project is using a private dataset, try to upload into github if the size permits. In all cases, the README file has to provide detailed information where the datasets are, how they    can be accessed, and how they are used.

○    If some part of your project  is too large for github (a very large model), upload it into your liondrive (google drive), and make the link to it accessible to                 [email protected]. Provide the information about the link in your            README file,  code comments, report, ...

Code

The code needs to be written by students and located and managed in the github classroom       repo.The code needs to be written competently, commented extensively, and be well-organized  into files, directories and subdirectories. The organization of the code should follow the                examples in the assignments - several python subdirectories with python files and a jupyter        notebook as the main UI file. At least one file needs to be a jupyter notebook file. Jupyter             notebook code needs to be functional so that it can be used by instructors to run your code, and thereby reproduce and verify some of the students’ results. The notebook needs to have plenty  of comments, and contain some of the key results including the figures with results. At the           highest level (root directory), there must be a README file, describing the following: instructions how to run the code, the name of the main jupyter notebook, description of key functions of         each file in the project, location  of the datasets, organization of directories and any other            supporting information.

The instructors do investigate if there is code available online for papers/publications suggested for projects, and/or for projects proposed by students. You are not allowed to reuse any code available online. You are expected to implement the final project in its entirety by yourself. You may reuse the code from homework assignments with proper references to it both inside your code and in the report.

If some parts of the code have similarity to or in some way originate from online resources, this needs to be commented extensively and original code must be referenced in multiple places -  both throughout your code and in the report.

If the rules are not followed, your code is likely to be declared as plagiarism.

Project due date will be announced in the courseworks (and/or github classroom).       Project proposals are not needed for papers which are on the instructors’ list of papers.

Grading components for project evaluation (tentative)

1.   OVERALL CONTENT (~25%):

a.   Review of literature

b.   Description of the problem

c.   Description of student’s implementation

d.   Flowcharts and block diagrams - student’s implementation

e.   Figures

f.    Quality of writing

2.   RESULTS  (~25%):

a.  Accuracy

b.  Training times

c.   Effort in optimizing results

d.   Figures

e.  Tables

3.   DISCUSSION  (~25%):

a.   Discussion of results

b.   Comparison with other paper(s)

c.   Discussion of problems faced

d.   Discussion of insights gained

4.   CODE  (~25%):

a.   README content

b.   File organization

c.   Code comments

d.   Code works

e.   Code is original, not plagiarized