In this assignment, you are challenged to explain a model. For this, you will research exisiting approaches and apply them to your model and interpret the results.
For the intermediate submission, please enter the group and dataset information. Coding is not yet necessary.
Team Name: TODO
Group Members
| Student ID | First Name | Last Name | Workload [%] | |
|---|---|---|---|---|
| TODO | TODO | TODO | TODO | TODO |
| TODO | TODO | TODO | TODO | TODO |
| TODO | TODO | TODO | TODO | TODO |
| TODO | TODO | TODO | TODO | TODO |
The submission is done with this repository. Make to push your code until the deadline.
The repository has to include the implementations of the picked approaches and the filled out report in this README.
- Sending us an email with the code is not necessary.
- Update the environment.yml file if you need additional libraries, otherwise the code is not executeable.
- Save your final executed notebook(s) as html (File > Download as > HTML) and add them to your repository.
Checkout this repo and change into the folder:
git clone https://github.com/jku-icg-classroom/xai_model_explanation_2024-<GROUP_NAME>.git
cd xai_model_explanation_2024-<GROUP_NAME>
Load the conda environment from the shared environment.yml file:
conda env create -f environment.yml
conda activate xai_model_explanation
Hint: For more information on Anaconda and enviroments take a look at the README in our tutorial repository.
Then launch Jupyter Lab:
jupyter lab
Alternatively, you can also work with binder, deepnote, colab, or any other service as long as the notebook runs in the standard Jupyter environment.
- Which model are you going to explain? What does it do? On which data is it used?
- From where did you get the model and the data used?
- Describe the model.
Find explainability approaches for the selected models. Repeat the section below for each approach to describe it.
- Briefly summarize the approach.
- Categorize this explainability approach according to the criteria by Hohman et al.
- Interpret the results here. How does it help to explain your model?
Write a brief summary reflecting on all approaches.
Provide the link to your recorded video of the presentation. This could, for instance, be a Google Drive link. Make sure it is actually sharable and test it, e.g., by trying to access it from an incognito browser tab.