South Big Data Hub Education & Workforce Working Group, 14 Oct 2021

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Education & Workforce WG: Building a School of Data Science and a Undergraduate Curriculum

Description

October 2021:
Building a School of Data Science and a Undergraduate Curriculum; Dr. Brian Wright; University of Virginia

A

Okay and uh I do uh dc data community, it was on the board there for people that don't know that it's a pretty big um kind of education, focused data community in washington, dc, okay. So um the context behind this, uh that's my research areas, there kind of data, science and education generally to be as broad as possible.

A

um The context behind this is for those that don't know at uva- and I know claudia is here. I saw that so she's heard much of this before so she's part of the school, but um we you know one of the first uh schools tree. I think the first school of data science in the country and as a result of that, the way that we went through kind of the process of building at least kind of retrofitting, to a certain extent, at least when it comes to the undergraduate.

A

Specifically before we started building the programs we said to ourselves because of the nature of which you know the obligation that we felt as the first school of data science is that we wanted to put more of a definition on what we meant by data science, whether we wanted that to be represented in the nature of the school and the nature of the programs and basically how we espouse to the larger society of academics and outward and industry professionals, what we meant by data science and that we're going to train our students inside that paradigm.

A

So and going through this. A lot of these may look familiar. These are all kind of your traditional.

A

uh I guess uh pipelines is what they are for data science and so what they are to a certain extent- and you know- there's there's more here- is that they're they're much more function than form right in terms of like academic depth, what we mean by the actual field of data science as compared to just the transactional elements associated with producing it.

A

So we took a step back much of this, led by uh rafael, alvarado and don brown, who are two professors in our school and we're kind of working on getting this out into the world.

A

But this is kind of our functional map in the way we think about kind of the theoretical foundations of the field of data science, and we are using this at least I am in terms of constructing the um undergraduates and then our phd programs as well, um uh in terms of how we, how we um put courses together, how we set goals for what the outcomes of those courses would be, and we were trying to balance uh these four elements so I'll go into some detail about what these are, but given the the pace of which we need to get through this I'll, probably use analogy, and so that's what I have here so to a certain extent, this is what we think of so I'll start up here at value when we think about the value portion of uh data science.

A

It's about business value. It's about this tension between open. You know the nature of open science and protected data, and then this larger kind of category of ethics. It's like the teacher and me coming out. I feel like I have to circle stuff, and then we have design, and so design is very much about kind of data, designing products, human-centered design, but also can be associated with data creation and how you curation, and how you design, databases right, so database design could go there.

A

So you think about this is like the creative elements or functional kind of elements associated with data analytics is the one that everybody knows pretty well right, optimization prediction: machine learning systems can be defined as kind of anything that is related to the actual production of uh of data science products. You know we think about. This is coding cloud computing uh things like that infrastructure, devops, and so some of this has seemed. You know kind of obvious.

A

I think in terms of those, especially people that are immersed into the field, but we didn't feel that there was like a really well-defined paradigm. That kind of laid all this out in terms of uh just how the field should be shaped, and so we wanted to have that in place inside the school. So we had kind of a reference point before we went forward and started building programs, we could say hey. We need to make sure to include all of these elements. So that's what we did.

A

uh Here's just again, some analogy here, associated with these different components. We call the four plus one model um and how we're thinking about data science. So maybe this helps kind of uh shape the landscape there a little bit more if needed, and then I have and I'll I'll send all these I'll slowly. These slides out, um I'm sharing with the community here, but there is uh a bit more depth in each one of these categories and uh we're working on kind of finalizing a series of papers.

A

Actually that we'll put out that kind of define this in in a bit in a you know in a more substantial way- and so um I was I was planning on getting into each one of these, but uh you know kind of at a very high level. We can see here kind of design is about this human machine interaction. It's really focused on kind of the human element associated with data science and how that comes comes to comes to fruition. The systems again is about kind of this infrastructure and architecture.

A

Analytics is kind of the most commonly understood. You know. Statistical methods in machine learning, things like that. There's tensions here associated with those as there are with the other categories, and then the value one like I said, is really centered on human value and kind of this disciplinary.

A

You know the discipline of ethics and how that interfaces with the field of data science, all along the data science life cycle by the way in all sorts of different constructs, but this is kind of what that um represents.

A

So again, that was pretty quick, but that's kind of the nature of what we're thinking about in these four different areas and then the practice is is simply the combination of all those things so putting those together uh and the balance of those not suggesting that every one of those categories is equally represented for every project, but every category should be represented in every project and typically is, and sometimes the balances between those vary depending on what the nature of that particular project is, and so the practice here too is is: has this kind of intrinsic element associated with domain knowledge and applying these things to real world problems?

A

So we have, you know, classes at the end that are associated with capstones or projects that are that fulfill kind of the requirement in this area.

A

uh So I did you know I uh the first slides I dismissed the pipelines and then you know here I'm bringing them back. uh So uh I created kind of this one, this four phases of the data science life cycle that uses the color coding associated with the different four and you can see how those kind of layout again. This is not it's not perfect. It's just a representation of the way that we think about it.

A

I think generally- and I use this in my courses to just lay out the landscape of how each one of these elements can represent in a more tangible way in terms of just pipelines.

A

You see here that the kind of the thing that maybe gets underestimated, but over estimated in this one or over emphasized, I guess, is a better way to say it is the value portion of the uh of the pipeline in terms of just up here on the front end and what we think is just actually creating a metric that can trace value and then tracing it once it's in deployment and then using these value ideas throughout the data science pipeline to make sure that we're you know kind of keeping keeping an eye on the ball there again.

A

I could go into lots of detail about this, but this is the idea is to take it now. You know our intellectual process was to go from. You know kind of what is the field? How do we see it? What are these critical components and then take that down to more of an applied way, which is what this is, so we put it into an actual pipeline as compared to what seems to be kind of the the shape of things.

A

Now they do in the opposite direction, say hey: what did we need to do and then go the other way? We wanted to kind of intellectually kind of define it largely and then go down um to the more tangible, practical things, and so we translated this into programs as well. So this is the minor in the undergraduate space here at uva. We don't have a major yet we're working on it, but we do have a minor and it has these five components.

A

If you um you, can google uva data science minor, when you pull it up? Actually, each one of these categories is defined by the four plus one model, so we have classes here, um as you can see in analytics, and then we have a systems course and a design or value, and then the last one is this domain course and then sorry, I don't know if the writing actually helps there. So here you can see it's kind of. We have a bit more uh really kind of two classes in systems.

A

So the way we think about it is kind of these systems. Elements are more foundational. You have to learn how to pro how to how to program right at the beginning, but uh and then and then move more into kind of advanced um topics relative to analytics, but it is just a minor. So it's a quick, you know it's a quick, quick, uh quick degree.

A

Just these five courses and the idea is to give them kind of a cross-section of the different flavors of the model um um to better to you know to understand the field. So that's the idea, and then here I use these little color coded boxes to see what the balance is in terms of just the courses we're offering.

A

So uh this is an example. We're extending this we're planning to extend it to the to the major as well. So this is all I need to put a big caveat in here. This is just an example, so we haven't decided any of this yet but again we're using this framework to make sure that we balance the different elements of data science as we see it, and so at the beginning.

A

Here, what we see is kind of a fairly balanced approach with maybe a bit more kind of focus on some of the foundational mathematical elements, but each one of these domains is represented and we want to make sure that that is consistent throughout the degree as students move up and just don't bias ourselves kind of towards analytics or systems and forget these critical elements associated with value and design, which sometimes can be the can be the case.

A

And so I guess I just put the I put that model everywhere and then this is upper level. Electives concentrations again just um you can see here we're kind of maintaining this balance, at least through this and again this is all theoretical. We have not decided any of this, um maintaining this balance kind of at the top portion and then uh there's a bit of a move towards uh analytics and systems to kind of foundationalize.

A

Some of these topics that students to to get more applied as they're kind of moving towards graduating, and then we finish with an immersion into which is to which is typical. I think, of these programs of data science programs to a more practical element where they do an applied project, whether that's a concentration in a one of what we're calling kind of collaborators, which is a relationship between the school of data science and another in another school, on campus on grounds or kind of doing an applied project.

A

So that's what that looks like, and then uh we also get. This is my course. So even the goal is to get kind of granular in terms of just all the way down to the course level. So these are individual topics on a week by week basis and because my this course again is kind of outside of a more you know, kind of scaffolded curriculum design where we're kind of giving the students a cross-sectional.

A

Look at the field. You can see that there's some blending here of topics with a good portion of actually you know practicing getting out there and getting used to using um some of these things in the in the future state. We see we'll probably have courses that will be much more biased towards in the individual topics. Much more biased towards specific uh domain areas would be. A value course would be, you know, dominated by value, driven topics, communications analytics.

A

So the idea is that you'll see less of this kind of homogeneous or less of this um mixed view and much more of a you know, a kind of consensus into one of the domain areas.

A

So we went from kind of you know the kind of this kind of this high level framework. We put it into a pipeline and do we do it also for broad course design and then we're also trying to just give it a cross just give it a snapshot even down to the individual course level.

A

In terms of the way things are being done and there's the so, I don't know any slides after this one, but um you know we're all kind of in the active uh process of building a school and the idea again we're uh very committed to this framework. I think we're actually going to construct the school in large part around that framework, especially having departments or people specialize in those areas. I feel like I'm up against time, so I'm going to stop.