Assignments

Assingment 4: Project report

The main idea of the report is to present the work that you have done as a group for the seminar. The length of the project report should be 5-6 pages. The latex, docx, and odt formats of the report can be found in the links [1,2]. You can modify the template based on your project outputs. The suggested sections of the report are listed below:

• Abstract: A paragraph that introduces the main idea of the project, methods, and results.
• Introduction: Frame your problem and the reasons that motivate you to solve this problem in the multimodal machine learning field. You can also briefly mention your results.
• Related work: Introduce the studies related to your project based on the papers that you read and emphasize the differences in your project.
• Methods: Describe the methods that you have employed to achieve the results, including dataset specifications, preprocessing steps, multimodal methods, evaluation metrics, etc.
• Results and discussion: Present your results by clearly defining your evaluation metrics. You can also compare your results with similar studies in the literature. Bear in mind that your results should be quantifiable. In this section, you should discuss the limitations of your approach and provide some educated guesses to mitigate the limitations.
• Reproducibility of the study: List the software stack that you have used for the project and provide a working link for your Gitlab repository.
• Conclusion: Briefly present what has been achieved and emphasize the key results. Additionally, you can introduce future studies of your work in different settings.
• Author contributions: Briefly describe the contributions of the team members to the whole project. If you are the only one in the project, you can omit this section.
• References: List the resources that you used for your project, including paper, reports, blogs, online repositories, etc. We recommend you to follow the IEEE reference guide to cite resources [3].

Assingment 3: Project presentation

The project presentation is an opportunity to introduce the current state of the project and get feedback from your peers. If your presentation date is within the first weeks of the schedule (especially, 05-06-2020 and 12-05-2020), you do not need to show the results for the full project. However, you can provide preliminary results from your project. The presentation should be 30 minutes long, including the QA session, and at least two team members should present it. Additionally, all team members should show their contributions by answering the questions at the end of the presentation. You can consider the following template for structuring your presentation:

• Project title, team member names, and course title
• Project objective and team motivation: You can briefly describe your aim and the motivation of the project.
• Brief literature review: You can summarize the approaches that exist in the literature to solve the same problem and mention how your approach differs from the state-of-art methods. This should include briefly describing the papers your project was mainly based upon.
• Multimodal pipeline: Provide a diagram that summarizes your multimodal pipeline, including the dataset preprocessing, implementation of the machine learning method, and the performance evaluation metrics.
• Results (or preliminary results): Provide figures, graphs, tables to present the results for your project, even if they are preliminary.
• Concluding remarks and future works: You should provide the conclusions of your current state of the project. You can also share some lessons learned which might be useful for the other students. If your project can be integrated into a different problem, you can provide some pointers.
• Some back-up slides: You can prepare back-up slides for the questions that you can expect from the audience.

Assingment 2: Project proposal

In this assignment, the teams should submit a brief description, ideally 1.5 pages long, of their project proposal. The topic of your project is shown in the table in the Seminar Groups section. Note that the (selected) topic of your project was assigned based on your paper preferences, i.e., the top three papers of your (or your group) choice.You can use the suggested template in latex or word format [1,2]. The proposal should include the followings:

• The title of your project and list of team members
• A paragraph that presents the objectives and motivation of the project
• A paragraph that introduces your multimodal machine learning pipeline
• A paragraph that introduces your proposed machine learning model (e.g., CNNs, MLPs, DQNs, AutoEncoders)
• A paragraph on expected outcomes of the project
• References

• Bear in mind that this seminar course is about multimodal machine learning. Therefore, you should process or introduce at least two modalities to perform machine learning methods.
• In your project, you might target to reproduce the same result by the papers that you read. The article might have a Github repository, or you may not have the same resources (e.g., GPUs) that the authors have. So, instead of proposing a replication study, you might think of re-purposing the model in the paper in a different task that uses different modality.
• In rare cases, we will allow to literature review as a project topic, especially if you do not have any group.

Assingment 1: Project team formation

In this assignment, students should form a team with a size of 2-3 members and choose their top three publications from the section: Papers for project topics and write a summary of the selected papers. The suggested length of a summary is 2-3 paragraphs.

• If you already have a group, the papers should be summarized by the group, i.e, one group with three papers. Then one of the group members should send summaries in any format (e.g., plain text, pdf) with group member's names and surnames to murat.kirtay@informatik.hu-berlin.de.
• If you do not have a group (or could not form a group) till the deadline, you should summarize three papers by yourself, i.e, one student with three papers. Then you should send your summaries in any format (e.g., plain text, pdf) to murat.kirtay@informatik.hu-berlin.de.