The grading is based on the final project related to training a deep learning model.
- The topic of the project is up to you. You can choose anything that interests you. If you don't have an idea for a project you can participate in Kaggle competitions. Example projects from last year:
- Distinguishing conspiracy theories from actual news,
- Classical music generation using MIDI datasets,
- Chatbot mimicking a particular person,
- Classification of chemical equations.
- The major part of the project should be the fine-tuning of a deep learning model.
- The project should be described in a short paper (~2 pages) which contains: the description of the task, the datasets you used, the process of creating the model, and the results.
- You can make the project alone or in pairs. If you make a project in pairs the scope of the project should be larger.
- 03.12.2022 - Send me an email with the project proposal. Even one sentence will suffice, e.g. "A classifier to recognize if it is raining outside based on a photo".
- 11.02.2023 - Send me an email with your final project, i.e. a short paper and the code you used to train the model.
- The most important part of the project is the process of creating the ML model. This includes but is not limited to: choosing the appropriate model and evaluation method, choosing or creating the datasets, and conducting experiments to train a model. I will also take into account the code quality, whether (and how) the model is deployed, and the final performance of the model.
- The performance of the model itself is the least important aspect of the project. You can still get the highest grade even if your ML model works terribly. ML is hard and it's difficult to predict whether a project will succeed or not. However, you should make an effort to make your model work as well as possible and this will be graded.
- For a passing grade (3) you need to meet formal requirements, i.e. train a model and write a short paper about your project.
- For a good grade (4) you need something extra, e.g. create your own dataset, implement the custom architecture, or "deploy" your model (I recommend Spaces which we will cover in the class).
- For a very good grade (5) your process of creating a model must be well thought out and well justified in your short paper.
Feel free to write me an email if you have any questions or if you are stuck and need some advice.