Description

Machine Learning
Mini-Project Instructions
Contents
1 Mini-Project Instructions 2
2 Suggested Reading 2
3 Project Group and Expected Workload 2
4 Data Sets and Methods Recommendations 2
5 Project Proposals 3
5.1 Project Proposal Part 1 3
5.2 Project Proposal Part 2 3
6 Project Report 4
6.1 Project Report Disposition and Content 4
6.1.1 Title 5
6.1.2 Abstract 5
6.1.3 Introduction (0.5 page) 5
6.1.4 Data and Methods (1 page) 6
6.1.5 Results (roughly 1.5-2 pages) 6
6.1.6 Conclusion (roughly 0.5-1 pages) 6
6.1.7 Bibliography 6
6.1.8 Acknowledgements (optional) 7
6.1.9 Appendicies (optional) 7
7 Project Presentation 7
8 Project Grading 8

1 Mini-Project Instructions
The last two weeks will focus on a course project where 2-3 students choose data and create a supervised machine-learning predictor for a real-world dataset.
Requirements for the projects are:
Your project should be a supervised learning project.
Real data should be used (see below for details). Choose a dataset you nd interesting!
Minimally, two methods (or two di erent neural net architectures) should be compared and evaluated.
Tip! Write the report to use it as an example of work you have done for future potential employers.
For PhD students: You can do a small project related to your research interest instead. Although, it should still be a 4-page paper output.
2 Suggested Reading
The project will be a small practical exercise in supervised machine learning. Suggested reading (before starting the project) is:
3 Project Group and Expected Workload
The project is expected to take 40 hours per student in the group. Hence, a 3-group project should be the equivalent of a 120-hour project.
Hint! If you work alone, try to choose a subject related to your master thesis project.
4 Data Sets and Methods Recommendations
We recommend that you nd a dataset you are interested in using yourself. If you have a hard time nding a dataset to use, there are a lot of available datasets (and problems) at:
The UCI Machine Learning repository: [here]
The machine learning competition site Kaggle: [here]
You should not use the following data sets in the project:
Titanic (R data set) mtcars (R data set)
5 Project Proposals
The project will be submitted multiple times during the course to even out the workload of the report and enable feedback.
5.1 Project Proposal Part 1
Students must turn in a half-page project and data description in the ICML paper format that can be found [here]. The ICML format is also available in Overleaf here: [here].
The project report should include all the group members’ names following the LaTeX template, i.e. your names should be as authors in the template.
The project proposal must include the following pieces. See the details on the di erent parts in Section 6.1.
1. Title
2. Abstract
3. Introduction
4. Data and Methods
In the Data and Methods Section, you only need to include the Data subsection and describe the data you will use.
5.2 Project Proposal Part 2
Part 2 contains additional information on the proposed method and some preliminary results. It should also x comments to the previous proposal 1. Students must submit a project proposal part 2 and data description in the ICML paper format (see above).
The Project Proposal 2 should include all the group members’ names following the LaTeX template, i.e. your names should be as authors in the template.
The project proposal must include the following pieces. See the details on the di erent parts in Section 6.1.
1. Title
2. Abstract
3. Introduction
4. Data and Methods
5. Results
In the Results section, you only need to include the preliminary result of one model/architecture.
6 Project Report
The requirements for the project report are:
1. The Project outcome is a report in the ICML paper format that can be found [here]. The ICML format is also available in Overleaf here: [here]
2. The project report should include all the group members’ names but follow the LaTeX template, i.e. your names should be as authors in the template.
3. The paper should consist of between three and a half (3.5) and four (4) pages, excluding references, acknowledgements and eventual appendices.
4. The report and project proposals should follow the disposition in Section 6.1.
5. The audience is students in the machine learning course, i.e., there is no need to go deep into details of what has been presented during the course. Although, if models not introduced in the course are used, you need to explain these models in more detail.
6. All Figures using colour should have a colour-blind-friendly colour palette. See here and here.
7. Before you submit the project, do a language check with a tool such as Grammarly (). A project with poor English (errors you would have spotted with a tool like Grammarly) will a ect your grade.
8. The nal report should be like a research paper, i.e., avoid bullet lists and get a good ow in the text.
9. This is an academic text. Hence, the claims you make should be backed by references.
6.1 Project Report Disposition and Content
The paper should include the following four sections and subsections:
1. Introduction
2. Data and Methods
(a) Data
(b) Methods
(c) Evaluation
3. Results
4. Conclusions
In addition, the report should also include a title, abstract, and a bibliography. The report can also include Acknowledgements and Appendices, but this is optional. Below are further information and requirements for the di erent parts.
6.1.1 Title
The title should describe the problem and be like an actual article title, i.e., don’t write “Mini-project:” or something similar in the title.
The project report and project proposals should have the same title.
6.1.2 Abstract
The abstract summarises your report in one paragraph of a maximum of 200 words. It should also be written in the present tense. We can write the abstract by writing one to two sentences on each of the following points:
1. An introduction to the topic.
2. Explanation of why the topic is important.
3. Your main research question, aim or problem.
4. Your research data, methods and models.
5. Your most important ndings.
6.1.3 Introduction (0.5 page)
In the introduction, you should:
1. Give a su cient background to understand the problem.
2. Description of the problem. The supervised problem should be well explained, including the target of the prediction (y) and the features (x) used. What variables are used here, and why is y relevant to predict using x?
3. Explanation of why the topic/prediction setting is important/relevant.
6.1.4 Data and Methods (1 page)
In the Data and Methods section, you should have (at least) the following three subsections:
1. Data
2. Methods
3. Evaluation
Data Describe the data you are using in your prediction model at a level so it is possible to replicate your analysis.
Methods The methods section should describe your methods so another student can replicate them in the course, independent of implementation language (e.g., R). The independence means that the hyperparameter settings should be explained and presented without reference to, for example, speci c R function’s default values.
If models with hyperparameters or architectures are used, there should be motivation. Why are they chosen, and how? You should describe the method used to set the hyperparameters.
Evaluation The evaluation should describe how you evaluate di erent models and what metrics you use. Hence, describe if and how you use training, validation and test sets and motivate the choices made. Then, present the metrics you use for evaluation and motivate these as well.
6.1.5 Results (roughly 1.5-2 pages)
Summarize the results of your models and compare your model results. Analyze the performance of your model and discuss the results. Try to summarize results in Tables and Figures to help the reader understand the results.
6.1.6 Conclusion (roughly 0.5-1 pages)
Connect your results back to the introduction. Did the method work as expected? Are the results good or bad? Discussion of problems and potential improvements. Also, include a paragraph on potential ethical/fairness issues (in light of the guest lecture).
6.1.7 Bibliography
You should use the correct reference systems. A tip is to use citet, citep, and bibtex. Using bibtex will simplify your future thesis work.
6.1.8 Acknowledgements (optional)
If you can let me use your project report as an example in the next course, please state that in the Acknowledgement. Just add the following sentence: “We grant our consent for the utilization of this project report as a reference material within the context of future editions of the course.” You can also add other people to thank.
6.1.9 Appendicies (optional)
Suppose there is additional material you want to include in the report that does not t. You are then allowed to use appendices. Only include appendencies you also refer to from the main text. Also, the reader should not need to read the appendices to understand the main text.
7 Project Presentation
Presentation details:
Each project needs to be presented in addition to submitting the mini-project report
The presentation should be high level, but su ciently detailed information should be readily available to facilitate answering questions from the audience
Within each session, about four groups will be presenting
For 1-2 person groups, the presentation should be 10 minutes
For three-person groups, the presentation should be 15 minutes
Afterwards, questions will be asked rst by other students and then by attending teachers.
Speci c presentation recommendations:
The rst slide needs to include the project title and group members’ names.
The chosen methods(s) should be explained and justi ed (you are not holding this presentation for a hypothetical customer who doesn’t care about the details of your methods). You should use
Big enough font size for text and gure labels to make it easy for the audience to read slides.
A good rule of thumb is to expect one slide to take 2 minutes to present.
The last/ nal slide needs to include your conclusion and the group members’ names.
Missing the project presentation
To be able to get VG on the project the presentation needs to be turned in the day before the presentation at 23.59 the latest.
8 Project Grading
Below are the criteria used when grading the mini-projects. Some general comments on grading are:
1. The more students, the higher the quality expected of the project, i.e. a better report is expected from a three-student report than a two-student report.
To pass the report (G), you should ful l the following criteria:
1. Turned in a correctly formatted report that follows the general outline of Section
6.
2. Show basic knowledge and understanding of the course’s core concepts by using concepts correctly.
3. Show an understanding of when certain methods should be used and how they should be used.
4. Use at least two (2) di erent methods (or architectures) and correctly compare them.
5. State what has been done in the report with clarity, good English and rigour so it is easy for a reader to understand and follow the paper.
6. Correctly use references in the report following the template guideline in Section
6.
To pass the mini-project with distinction (VG), the additional criteria also apply:
1. Show deep knowledge and understanding of the core concepts and how to adapt them well to a new situation.
2. Connect the analysis in the report with other areas in statistics or machine learning or previous courses taken in the master’s program, i.e. not just repeat what has been done in previous labs.