Create a list of user stories

[cwmeijer]

As a <developer/domain scientist/ML engineer/...> I want in order to .

Link to the list of user stories: https://nlesc.sharepoint.com/:w:/s/DIANNA/EdoppPpmTTBKh0Mq-VdmvM8B_VeTFRzznZ3op-wo1cuBPA?e=qtGO5M

[elboyran]

Blog post about user stories in Agile development.

[elboyran]

Unfortunately, no responses from Sonja (as a link to an interested PI) nor Elisabeth (SSI-ML). Yang will remind Sonja, I will write one on Elisabeth's behalf and/or maybe Leon can also ask if soon to meet in the SSI-ML context?

[elboyran]

An example user story based on my understanding of the wishes of Elisabeth Lodder from AMC:

As a domain scientist (Elisabeth) I want to have an XAI tool which works on time series in order to verify (from a medical expert view point) an DNN-based detection of a specific medical syndrome from EEG signals.

[elboyran]

Generated from suggestions by Nico:

As a domain scientist (a chemist) I want to have an XAI tool which works on graph structures (like molecules) in order to understand the importance of a functional group in the molecular structure as predicted by a graph DNN.

[cwmeijer]

These are helpful stories! I had not thought about the graphs to be honest. The stories tend to be a bit general though which would make it hard to convert them into tasks later. I'll make some suggestions that would make them more specific: Elisabeth: As a medical domain scientist I want to see heat maps of my time series data (EEG signals) wrt classes in order to an DNN-based detection of a specific medical syndrome. As a medical domain scientist I want to edit (draw?) my time series data (EEG signals) in order to explore the change in behavior of an DNN-based detection of a specific medical syndrome. As a medical domain scientist I want to zoom in and out in my time series heat maps in order to inspect details. As a medical domain scientist I want to export my heat maps to a convenient format like csv/json/... in order to analyze the data further with my own tools.

Nico: As a chemistry domain scientist I want to view my network and have edges thickness / color coded wrt importance on classification in order to understand the reasons of the prediction by a graph DNN. As a chemistry domain scientist I want to filter my network visualization using filters in order to inspect the map with less clutter. Note, I didn't understand the whole functional group thing so I cut it out ;-)

Do you think this is what they want? What else do they want? Obviously they want the tool to do all of the above and more, but what is really important to them?

[elboyran]

I don't have experience with user stories, so yours look fine. What I was aiming for was the high level scientific questions they seem to want to answer. I guess it's then up to us to convert those into technical requirements and your stories seem to be doing that ;-). Nico gave the functional group as a concrete example, so maybe we can ask him if these definitions of user stories make sense? I can also forward your stories to Elisabeth & Simona for confirmation they make sense to them too?

[cwmeijer]

Yes, ask them to look at these and think about some more that are specific. After that, it is important to know from them which ones are really important and which ones are less important but convenient etc. Otherwise I'm sure they will just agree with whatever complete list we will show to them. This would make it very hard to make any minimal viable product. We should aim for a list that is so complete, that we only need to build functionality that is directly necessary for at least one user story.

I would expect that a feature like the ability to draw on eeg data to alter it (point 2), would be something fairly low on the priority list. It is there also for inspiration to create more user stories like it. But that is up to you, as a product owner, to figure out, with help from our stakeholders, and your team of course :-)

[elboyran]

Feedback by Simona (Post-doc for Elisabeth):

An alternative can be the following:

As a domain scientist, I want to have an XAI tool that works on time series in order to identify the importance of each data point of the time series given by the DNN-based prediction model for the detection of a specific medical syndrome.

I have done this for the LQTS project, using a modified version of the CAM-Grad technique. If there are alternative methods that can be used to understand how important each data point is, we can compare the results of the different methods. The importance can be a score, and it can be afterward color-coded and visualized on the time series.

[elboyran]

Generated by me from a finalist proposal for Loretnz eScience workshop:

As a climate scientist (Dim) I want to have an XAI tool which works on time series climate data for seasonal forecasts in order to gain trust and understand the sources of predictability

[cwmeijer]

Does Simona maybe have a notebook or report about the CAM-grad technique and LQTS project? If we could see what she did there and what specifically she looked at - so here whole procedure basically - that would be very helpful while prioritizing possible dianna functionalities.

[elboyran]

Generated by me from a finalist proposal for Loretnz eScience workshop:

As a climate scientist (Dim) I want to have an XAI tool which works on time series climate data for seasonal forecasts in order to gain trust and understand the sources of predictability

[elboyran]

Generated from suggestions by Nico:

As a domain scientist (a chemist) I want to have an XAI tool which works on graph structures (like molecules) in order to understand the importance of a functional group in the molecular structure as predicted by a graph DNN.

Update:

As a computational biologist working on protein-protein interactions, I want a XAI tool on graph structures that helps me understanding the importance of atoms/residues as well as interactions between pairs of atoms/residues in the forming of a stable protein-protein interface.