Project

The project is performed in groups of 4-5 students in the period December 4 - January 18. To make the groups as homogeneous as possible in terms of ambition level for this course, we will divide you into groups dependent on grade performance on Assignment 1. PhD students will be placed in separate groups.

The task of the project is to reproduce the results presented in a published scientific article, describe the article orally and in written form to your peer students, and argue for and against the method presented in the article. From this you will learn how to read scientific articles, how to implement and use a particular method, how to argue for and against a method, and how to adapt the presentation of a method to different target groups (i.e., adapt the presentation of the method in the article - targeted to active researchers in Machine Learning - so that it is understandable to first year Master students in Machine Learning).

Below, you will later find lists of scientific articles; Each teacher has provided their own list, and will take responsibility for the supervision of the projects on their papers. 

• Hedvig's list
• Carl Henrik's list
• Jens's list

Supervision

The group should plan and carry out the project autonomously. However, the teacher who listed the article you work with is available to answer questions, should you get stuck.

All teachers are away (at the Machine Learning conference NIPS) but are reading email December 5-13, and on holiday December 24-January 1.

Grading

The requirements for pass (i.e., E or higher) are

• an implementation of the method and recreation of the experiments, according to the instructions below,
• a written report according to the instructions below,
• an oral presentation according to the instructions below.

Higher grades (A-D) are set based on

• how well you are able to repeat the results in the article (or alternatively, how well you argue for deviations between the results in the article and your own results),
• how well you are able to argue for and against the method in the article, and
• how well you are able to present the method in the article, making it understandable to your peer students. Both the written and oral presentation are taken into account.

Implementation

Each group should implement the method as described in the paper. You may use any programming language. You should implement the entire algorithm "from scratch" to fully understand all details of it. there are exceptions to this, such as complex optimization schemes, which are too time consuming to implement - these exceptions are listed in connection to each paper.

You are not allowed to contact the authors and ask for their source code.

You will then perform the same experiments as those described in the article. Ideally, you should get the same results. If this is not the case, you must make an argument about possible reasons, and prove your argument by small experiments which you design and carry out yourself. A negative reproduction result together with a good argument gives an equally high grade as a positive reproduction result.

All group members must participate actively in the implementation, i.e., write code. Each piece of code that you write should have a comment stating who contributed to that piece of code. You do not have to submit the code for assessment, but be prepared to show it upon request. (For more information about code plagiarism, se General > Honor code in the menu to the left.)

Written Report

The article, the re-implementation, and your results are presented in a written report, to be sent to Hedvig Kjellström via email in pdf format before the oral presentation on January 19. The report should be at most 10 pages, with 12 pt font and about 2.5 cm margins, including references, images, and tables. In addition to the 10 pages, the report should have a cover page with title, group number, author list and (optional) abstract. The report should be written in (to a reasonable level) grammatically correct English.

In the report you should first describe the article on such a level of detail that your peer students in this course understand the method, and so that it is clear to the reader that you understand the method too.

You should then present your re-implementation of the method, and your reproduction of the results, again on such a level that your peer students understand what you have done, and so that it is clear to the reader what results you got and if, how, and why they deviate from the results presented in the original article.

Finally you should argue for and against the method, possibly suggesting improvements.

All statements made in the report (e.g., "method X is better than method Y") should be supported by either a reference to the original paper or report where the statement was made, or if the statement originates from you, you should explain why this statement is true.

A technically correct, well organized report with good language and a clear line of argument will receive a high grade. Missed hand-in deadline, violations of the length and formatting requirements, as well as statements not supported by references will have a heavy negative effect on the grade.

All group members must participate actively in the writing of the report. By adding a group member to the author list of the report, you certify that this person has written at least one section of the report.

Oral Presentation

You will also present your project to the other students of the course in a short oral presentation on January 18 at 14-18 in hall TBD.

Each presentation is 10 minutes long and should follow the this format:

• The aim of the article - what problem is addressed?
• The method
• The results that they get in the article
• The results that you get
• If there are differences - what are the reasons?
• Arguments for and against the method

A clear, well organized presentation will receive a high grade. We can assure you that you will not get lower grades if you are nervous and make mistakes because of that. However, do not try to cram too much into the presentation - violations of the 10 minute time limit will have a negative effect on the grade.

Slides are optional, but if you have them they should be in pptx or pdf format. Bring the slides to the session on a memory stick, so that they can be uploaded to one computer in the beginning of the session.

All group members must participate actively in the oral presentation, i.e., talk for at least a minute.