Skip to main content
Loading Explore guides…
Data Visualization in Science · Acedra
Home
Explore
Computer Science
Data Visualization in Science
Sign in to save
Chat
If YouTube asks you to sign in, open the video on YouTube.
Watch on YouTube
Reload player
Transcript
77 segments
Copy
Hide video
0:01
- Thank you for coming. My name is Alyssa Goodman, and I am the co-director of science program here at the Radcliffe Institute for Advanced Study. I also have to admit to being an astronomy professor. And so you might see a slight extra dose of astronomy, just a tiny extra dose of astronomy today, but really tiny. And I'm really very, very excited about this program. And so the program, as you know, is called Next in Data Visualization. And it's part of a bigger program at Radcliffe
0:29
that happens once a semester called Next in Science. And I have to give credit to my colleague in physics, John [? Hughes, ?] who used to have my role here, who came up with this program with the idea that listening to old people is one thing, but listening to young people is even better. And so the young people that you'll hear from today are the four across the middle of the screen here. And I'm going to introduce them one at a time as they begin to speak.
0:54
And, right now, I thought I would just introduce the overall program. But, before I do that, let me just tell you that our general strategy is we have these four speakers. Two of them are going to give you talks before we have a very short break and then two more after the short break. And then we have a question and answer-- not question and answer, sorry, a panel discussion at the end among all five of us, but you may ask questions of the individual speakers after their talk.
1:18
So we have about five minutes for questions after each talk. So, if you have a question for a particular speaker, please ask that question if you can during their question and answer sections. And then there is a reception at the end if I forgot to mention that. So please join us for that. Those of you who do Twitter, the Twitter handles of each speaker are up here now. There's a tweet telling you what the online livestream of this event is in case you want to share it with your friends.
1:44
And the hashtag for the event is #RadDataViz. It's pretty cool. I didn't even come-- I can't take credit. I love that, but yeah. OK, so who remembers doing something like this or having your child do something like that? So raise your hand if you remember this exercise with the apples in school. No, because you're old too old.
2:09
So these days, OK, kids, especially in Montessori-like schools and other very hands-on schools, are asked to think about statistics very early on. And the way that they think about it usually is by being asked what's your favorite kind of apple-- red, yellow, or green. This poor kid is counting. He's putting his apple on the column that he likes. And this is a particularly difficult challenge because they have two different kinds of red apples. So it's like Delicious apples, Macintosh apples,
2:40
or Granny Smith apples. But, anyway, this is a very, very direct bar graph. The next thing that happens is they get a little more complicated. And then they said, you know, who likes which bar graph. And then there's a list of each person-- each kind of apple. There's a list of each person and what kind of apple they like. Now, in the world that you're going to be hearing about today, that list is called metadata, OK? That's attached to the data, which is the number of people in each column.
3:08
And that doesn't look very much like a bar graph anymore, but it is a bar graph. And, you know, this view looks a little bit more like a bar graph. And there's a statistical analysis here on the side. You see the most green, the least yellow. It's a statistical analysis. And the metadata has now been embedded in the graphic, OK, with the names on the apples. And I'd say this is actually a pretty good graph. It's probably better than the one on the right here, which comes from a worksheet.
3:36
And you could replace the E-E by an I and know what many people in high school and middle school call worksheets these days. I won't say that out loud, but the reason is that they don't present usually a very high-quality learning experience. And, if you look back at this kind of very direct data visualization and this kind of data visualization on the right-hand side here where you have these bars, and you have to try to interpret what those fruits are,
4:01
and you have to read the tiny numbers off the side and figure out who liked what and how you count it, it gets worse and worse. And, as graphics get more complicated and hard to read, people lose interest, and they think that they're confusing. And they can't understand them, OK? So the one on the other side, on the left, is actually an interactive, which is kind of the equivalent of move your apple to the column that you like where you drag the apple to what you like.
4:27
And that brings in interactivity, which you'll hear about a lot today. And the reason that I chose these examples was to get you to the point, not of thinking about apples, but of thinking about how there's a huge disconnect between the kind of visualization and statistical appreciation that people have when they're five years old in a Montessori school and when they're much older than that and have to understand interactive graphics like this that are arguably about the most important problem
4:53
facing humanity, which is climate change, that you'll hear about from at least one of the speakers later. And you'll see that that's not just a movie. That's actually an interactive where I can move the Earth around, and I can compare here on the left. This is the difference between the average temperature of the earth over a century and the average between 1959 and 1988. And then here's the average for the next period, 1988 to 2017. And, without even reading this color bar,
5:21
you can pretty easily tell that, on average, the Earth has tremendously warmed between these two periods, OK? This is a very nicely done graphic where there's a proper color scale that you'll hear a lot about the perception involved with those from some of our speakers later. But my only point here to get started is that, when you think about these different kinds of apple visualizations, as simple as they are, these rules that are available on a--
5:49
not rules, sorry, but questions about visualization that you can ask of professional scientists, which is who this list was designed for, apply as early as visualizing apples when you're little, OK? So they say things like who is your audience. Are you trying to explore? Are you trying to explain? Are there important categories? What color are the apples? Are there patterns you're looking for in the data? Do you want to make a prediction or uncertainty? What are the dimensions?
6:16
And remember dimensions don't have to be spatial. What about abstraction and accuracy? So you can see the least abstract is put an actual apple down. The most abstract is make a bar graph, and you have to look up what the number of apples is, OK? What about context and scale? I think these get more and more obvious. The metadata is the names. And display modes is something you'll hear a lot about today. Is somebody going to have a static piece of paper? Are they going to have a three-dimensional
6:44
visualization? Are they going to have a way to interact with it? Are they going to have a very physical, three-dimensional representation that they could interact with? And all of these affect the choices that people make in visualization. And the reason this is such a timely topic right now is because technology has brought us so many new data sources and new ways to display data. And you'll hear about both of those from our speakers today. And this list is available, like I said, as a handout
7:12
out in the back of the room along with another handout that our first speaker, Michelle Borkin, will tell you about. So I'm going to tell you about Michelle in just a second. But what I want to do is I just want to say that the speakers are approximately grouped in talking about exploring data, the first two, the ones before the break, and explaining data, OK, not strictly. A lot of them want to do both. And at the discussion-- they don't even know this. I'm springing this on them.
7:37
OK, but the discussion I want to have at the end is actually about going back and forth, OK, between exploring data and explaining data and whether the tools of the future will do more and a better job of that. OK, so now back to our first speaker, Michelle Borkin, who will come up here in just a minute. And let me just give you a little bit of background about Michelle. Michelle is an assistant professor now at Northeastern University. And she's co-director of a consortium called the Northeast
8:07
University Visualization Consortium, which was started recently. And we know Michelle very well at Harvard because she got her undergraduate degree here and her graduate degree here. And she worked with Hanspeter Pfister, who's right here and somebody I know very well, on work that she's about to tell you about in her talk, which is called data visualization across disciplines. And, as you will see, she's an expert on human computer interaction and fields
8:37
ranging from astrophysics to medical imaging. And she'll also touch on topics involving perception that we'll hear more about in the second part of today. So come, Michelle, please. [APPLAUSE]
8:55
- Thanks again, Alyssa, for the lovely introduction. It really is an honor and exciting to be back on campus. Now I spend my time on the other side of the river. And I'm really excited to talk with you about sort of what's next in data visualization, obviously biased in this talk by what I'm doing next and working on, but it does relate to what everyone else is going to be talking about today. So a little more about my background, I majored--
9:21
concentrated in astronomy and physics here as an undergrad. I'm an avid amateur astronomer. And my approach to sort of how I got into data visualization was being a domain scientist with a need to visualize my data. So here I am at the Green Bank radio telescope in West Virginia and have a secret subspecialty and interest in radio astronomy and star formation. But it was during my senior thesis and working with Alyssa Goodman as my undergraduate advisor
9:50
that I had the challenge of having to visualize my data. I had tons of multi-dimensional cubes and tables and statistics. And it was the process of being forced through my undergraduate thesis to visualize this data that I retracted my applications from graduate school programs in astronomy and said, whoa, I like visualization. Let me explore this. I was at the Initiative in Innovative Computing for a couple of years as a research assistant.
10:16
And that's where I said, OK, I like this computational science thing. I really like data science. I like data visualization. So I came to the engineering school here and was in applied physics, but with a strong emphasis on data visualization in HCI. And, somehow, even though I started in computational fluid dynamics, by the end, my interests and novel contributions were in data visualization. I was sort of off and running in that direction, graduated. I just did a postdoc with Tamara Munzner.
10:47
And now I'm faculty at Northeastern. So this is the sort of timeline of me going from domain scientist to data visualization, interdisciplinary collaborator. Now, when I show this slide, I think it's important to say these are only two facets of what was going on in my life at the time. You see my research. You sort of see where I was in job title. I also always like to point out I got married about this point when I was in graduate school. And here is a photo from our wedding.
11:14
And now my husband Moli and I are the proud parents of a baby girl. So this is what I was doing the rest of my time when I wasn't playing with data visualization. But we're going to head back up to this point, sort of that moment where I had the, aha, I love data viz. What is this? I must do more of it. To make sure we're all on the same page on the first talk today, what is visualization? So the actual formal definition of data visualization is the visual representation of data
11:43
to reinforce human cognition or, more simply put, things that look like these. We're talking bar charts and line charts, scatter plots and maps, diagrams. These are all visualizations. But I'm going to augment this definition to say it is the static or interactive visual representation of abstract or spatial data. So these are all static visualizations. Interactive would be anyone checking the weather on their phone, stock prices, read those beautiful New York Times
12:11
articles with the interactive visualizations. That's interactive. And then static we have here. Abstract and spatial, abstract is your quantitative data-- bar charts, line charts scatter plots. Spatial is more-- you know, it's actual the world we live in. It's things like maps, diagrams, 3D renderings from the sciences, globes, any celestial astronomy, anything that's spatial. So this is what we're talking about today. OK, but why do we need visualization? My favorite main justification is
12:42
we have a lot of data we need to look at. And it's coming from all sorts of sources-- scientific simulation, text digitization, gene sequencing, financial business data. And we have so much data we really don't know what to do with it. And visualization is key to helping address some of the challenges-- so issues of scalability, sort of complexity reduction-- so we can take this huge amount of data and interpret it in a single graph-- and also keeping humans in the loop.
13:10
And this is becoming a growing need of understanding the algorithms behind our data crunching or understanding how our data is stored and where it is. So let me give you an example of one of these sort of big data challenges. And this was one of my favorites from when I was doing my PhD. This is the Cassiopeia A supernova remnant. It's one of the youngest in our galaxy, about 330 years old. And a colleague of mine, Tracey DeLaney, who was at MIT at the time came to me
13:37
with eight years of gorgeous data from Hubble, Spitzer, a lot of Chandra X-ray Telescope data, ground-based information. It was cubes. It was images. It was catalogs. And she didn't know how to put it all together. So what you're looking at is the beautiful rendering that was made for the press release. I lovingly call this the Hollywood version of her data or the cover of Vogue magazine. This is not what she actually did the science on. This was the first draft I made with her.
14:05
And she instantly saw this and could see the sphere was not symmetric, and there are these shells. And the jets were coming out in different ways. And it completely changed how she and others thought about the supernova remnant, as well supernovas in general. But, little did Tracey realize, I used software to do the registration, the segmentation, and the visualization that was developed at Brigham and Women's Hospital for surgical planning. And this is the classic interdisciplinary case
14:35
where another field developed this incredible tool that could help her look at her data that and analyze it. So I don't have time to tell you about all my projects, but I've spent the last 10 years sort of working across these different disciplines with physics and astronomy and chemistry and biology and geoscience and, more recently, computer scientists and cognitive psychologists and vision experts, digital humanities, to take tools and techniques from one discipline and sort
15:03
of cross-pollinate into another and work on collaborative projects together to sort of solve mutual challenges and problems, specifically using data visualization. So one area-- we're talking about sort of Next in Data Visualization. Before I dive into the main part of my talk, I wanted to give a hint at one area that I've become very interested in recently, which is how many of us actually read the news on our computers or browse the weather or browser data. We use our phones.
15:32
And so this is a growing challenge now in the visualization and the HCI communities is how do I take these visualizations and have them on my phone or my smartwatch or maybe a table or a tablet. And my student Michail Schwab and I have been looking into this and collaborating with experts in human-computer interaction to make interactive visualizations more intuitive and better. So this is motivated by a couple projects. These are digital humanities projects of ours.
16:00
And it had to do with effective timeline design. And when we went-- we made the timeline, and we sort of figured that out. But then I was like, how do we interact with the timeline? And we were overwhelmed. We didn't know like which zoom method and which pan method. And is someone looking at it on a computer? Or is it a phone where you have gestures and different fingers? And so we did this massive evaluation. We're going to present the results next month at CHI,
16:25
but, needless to say, there are certain techniques that are more effective and easy to use than others. We publicly released a library to support this. And a big part is design considerations for mobile platforms and mobile visualization. If you don't believe me that you interact with your phone a lot, in particular panning and zooming-- so, in a typical day, if you're panning on your phone, on average, you're going to pan the distance of about 25 meters, which means in a year,
16:53
if you're using your phone every day, you're going to basically pan the height of Mount Everest. And so the challenge here is, if you have an interaction that saves you 10% or 15% efficiency, it's quite literally the difference between scaling Mount Everest or not. Now the other area I'm going to talk about, which is a future area and a hot topic in visualization, is health care. And how can we use effective visualization techniques to help with diagnostics in health care applications?
17:22
So what you're looking at here, this is my dissertation research. And we're going to talk about heart diagnostics for a couple of minutes. These are your coronary arteries. So your heart is about the size of your hands. And these arteries wrap around your heart and supply blood to it. And if you get a blockage or a rupture in these arteries, you're going to have a heart attack and probably die. So I joined a team of physicists and visualization experts here at Harvard in my dissertation work
17:50
to say how can we present data to the doctors in the most effective way so that they can figure out where these blockages or diseased tissue are located. So this is sort of the conventional representation the doctors were using when I joined the team. I came in, did a huge iterative design process and a user study, and came up with this version on the right. It's a flattened-out projection of the 3D representation. I changed the color, and I sort of ordered this in a new way.
18:21
The doctors appreciated the idea or the motive of 2D. With 3D, if you're actually scrubbed up, and you're trying to do surgery on the patient, you can't interact with a 3D thing. And so 2D was easier to interact with and also, hopefully, faster for identifying these diseased regions as a spatial search task. In these, at the bottom of the color scale, is the diseased regions. So I went ahead and built a prototype tool with my collaborators. Here is looking at one patient, a simulation of them
18:53
at rest and in a stress test, a very diseased patient. [INAUDIBLE] built this tool. You can browse. You can look at the trees. We have a focus and context view. So I did this design study, built the tool, gave it to the doctors, walked away, and, a couple months later, come back, and no one is using it. And I was disappointed. I was like, well, wait. Why aren't you using this? And they were not-- they weren't convinced that the 3D was-- they wanted that one. They thought it was better.
19:22
It was anatomically-spatially correct. And they had not bought into the 2D. So I did a user study and decided to evaluate the accuracy of diagnostics with the conventional 3D and the new 2D. I recruited 21 Harvard medical students and did an extensive quantitative lab study. So what were the results? For 3D, there was a big difference on how many diseased regions were identified. And, on average, it was about 39% versus 62% in 2D. But I was interested in color.
19:58
For anyone who's in the viz community, we all sort of have this body of knowledge that rainbow maps are bad. We can't perceive differences in data very well. There's no luminosity variation. So I decide, what happens if we change the color map? So that was a condition in my experiment. The accuracy jumped up to 91% on average. And then, finally, when I had this result, the doctors were willing to chat with me about the idea of going back to the 2D representation
20:26
and reconsidering their color choices. You'll hear a lot more wonderful stuff about color and perception in general in Danielle's talk later. Now this was heart. And this led me into my current research. And now, working with my student Aditeya Pandey, we've leveled up in the game. And now we're focusing on arteries and blood flow in the brain. This is so much harder. You can look in this MRI scan image here that it's just as googly-goop of mess. And now the question is a doctor needs
20:54
to identify a clot or a blockage or a stroke. How can they find that in this mess of an image? So here's an illustration, just to give you a little context of what you're looking at, with the cerebral arteries and the carotids. This was our first draft. So our first draft was let's do 3D and do a really good job of 3D. During pilot testing, the same thing happened. The doctors got lost. They had a lot of trouble. So over a long, iterative design process, we made this 2D representation.
21:25
And we've just submitted this. And we just did an evaluation. And yes, it works better. The 2D, the doctors are able to diagnose a patient more effectively than 3D. But what was the secret sauce here? The secret sauce was an interdisciplinary collaboration. And let me explain why. So, if a doctor looks at this, they're going to look at this branch and say, oh, there's a narrowing. There's a stenosis. And the blood flow is kind of out of whack in this representation.
21:50
Let me take a look at this branch. And they sort of follow along the branch of the artery. It turns out what they're doing is the network science or topology task of path following. And so the big breakthrough in this project was network science and graph theory. And, collaborating with members of the Network Science Institute at Northeastern, we've creating novel definition of the cerebral arteries as a network and informed our visualization, our design, and our analytics
22:17
by talking with our neighbors on the other side of campus to think about analytics and visualization aspects from network science. In case you're curious, the cerebral-- the arteries are a directed binary tree with an undirected weighted cycle and directed edges at the bottom. So, in my lab, we work in a lot of these decision support system cases where we have heart. We do some brain imaging, glaucoma diagnostics, as well as brain tumor diagnostics-- so health care and visualization.
22:49
But let's take a step back because both of these studies imply the 2D representation is great for these diagnostics. And I don't want you to think 3D isn't important. 3D is essential for the diagnostics and planning for follow-up treatment. Once the doctor finds the disease, they need to understand the heart function, or they need to understand the blood flow in the brain and plan for a surgery. Are they going to do a stent insertion? Or are they just going to do medication?
23:14
And, to make that possible, you have to take the 2D and the 3D and make them linked. You have to have them communicate with each other. So let's take a step back here. Here is the 3D visualization. This is the heart case. They care about this for surgical planning and stents. They want the original raw data. Here's a CT scan of the heart. They want the abstract representation so they know how to diagnose the patients. But then I didn't even talk about things like heart function.
23:41
And they have all sorts of other information. And what they really want to be able to do is have all of these in a single view, have them connected, and be able to go back and forth. Guess what? The astronomers have the same problem, not life-threatening, thankfully, but it's the same kind of thing. And here's the L1448 star-forming region. They have the 3D rendering. They want the 2D original radio telescope data, the abstraction that shows the hierarchical structure of the cloud.
24:08
And they care about things like comparing to simulations. And this is sort of ubiquitous and common across the imaging sciences where you have these different multi-dimensional views, and you want them together. And, specifically, you want to be able to link. And here I've highlighted in red the cube in the 3D view. And I, as a user, want to select something in 3D and have it, the same data, propagated and highlighted in all my plots. And this is really common.
24:35
And we've talked with people in astronomy and physics, biomedical imaging, chemistry, geosciences. They all sort of have this same situation. And then the crux here is gluing. So you can visually have that red linking. But you can't do that unless the data is connected together. And I call it gluing because it's like take this [INAUDIBLE] and say, oh, this one is in this projection spatially. I'm going to [INAUDIBLE] translation and glue it here. But it could be non-spatial.
25:03
I could have temperature measured in all of these, but they're different regions of space. And you still want it together. And so this gluing is really data management or schema mapping to be precise. And it is challenging, as I said, where you might have some spatial data and some non-spatial data, if you look at the variables on the side, that you have different things. So this had been like the Holy Grail for myself and the collaborators,
25:26
including Alyssa Goodman, Tom Robitaille, Chris Beaumont down the street that way at the Harvard Smithsonian Center for Astrophysics. And, after a number of years, this was our first sort of of working prototype. And it was this aha moment when we were able to load a very large image or a cube into a viewer, allow a user to pan and zoom, to select points here on the scatter plot, color diagram, for those who aren't in the field, and see it link across back and forth. And so this is called glue.
26:01
This was the old, original version. This is the updated, slightly beautified version. And this is completely open source in Python. I will have flyers in the back of the room. So, if anyone is interested, you can get some more information about it. And the goal here was, even though our first and true love is astronomy and physics and supporting it-- and I should note that we are an official visualization tool kit working with NASA for data from the James Webb Space
26:30
Telescope-- we built this to be interdisciplinary for all those reasons that I showed you before. So we can help scientists across all these different domains who have the same challenge and the same problem. So, for example-- so now I've animated this. And you can see an astronomer going through and interacting with the data. We do this in 3D. This is a data set with earthquake depths and location. And so not only can you select data in the 2D representations, the scatter plots
26:59
or the histograms, but you can also interact with the 3D representation and select points here. We are working with multiple medical collaborators in the Boston area in work in development. So it can read in medical imaging files of different formats and have multiple data sets rendered to volume, rendered in the same view. And this brushes and links across all these other windows. And, hey, if GUIs are not your thing, you can type everything in Python. And you can completely control the GUI
27:30
and what's going on with an IPython Terminal window. We have a lot of work going on right now with Jupyter Notebooks, JupyterLabs, all through Python. And a lot of this was not just to give the user control of their data, but it's also reproducibility and being able to sort of put this in hard code and take your user interface and extract it out. And then this is a shout out to Catherine Zucker for getting this to me last night, some of the newest work in glue, which
27:58
is going beyond what you can do in glue itself. And now it's let's take the best tools in visualizations out there and bring them in and brush and link across software and across platforms. This is the WorldWide Telescope. And they have a beautiful web API for looking at astronomy visualizations and data, both for communication and exploration. And here you can see someone interacting with a 3D model of the galaxy, and they've brushed and linked
28:26
their green points on the galaxy from their data across systems into WorldWide Telescope. And this is a future direction we're moving in. And, finally, the stuff I was showing you earlier with the medical diagnostics, we are actively developing this in glue and expanding the tree and network visualization capability. So, across all these different disciplines I've shown you and brushing and linking in different ways, this is sort of our hope is to continue
28:54
to expand the capabilities of glue and similar software to be interdisciplinary applications and help scientists across different domains. So, with that, I'll wrap up. And I can take a few questions now. And, otherwise, I'm excited for the panel later. Thank you. [APPLAUSE]
29:18
- That was so awesome. Thank you. As probably the least knowledgeable person in the room, I tell Alyssa, when I look at a visualization, I go like this. [LAUGHTER] So my very naive question is, when you have that type of medical imagery in a 3D, would it be useful to superimpose a heart, a three-dimensional heart or a brain, something that brings to the real novice, everyday analytics level to make it feel more real to me? - Real or-- there's two things that would do.
29:48
So the idea is, if you have like a 3D heart or a 2D-- or a brain, can it help someone understand what's going on, especially so they don't get lost because this happens all the time. And I think yes. And, specifically in radiology, they're very well trained to look at the brain in certain ways. And we have it very hard-coded to sort of match certain views so they don't get lost. But we're actively thinking about-- we call it sort of having abstract representations, but with context.
30:19
And I think that's an important concept is how do you provide enough context so they can understand something that's abstract. Or even if it's not, even if it's the raw imaging data, how do you give some context? So I think your point is spot on and a really important area to keep working on. Yeah? - Hi, thank you very much for the presentation. I have a quick question. So you talk a lot about how you can make new visualizations that are objectively very good.
30:47
And there is another part of the job that is to use a language, design language, grammar, a visual grammar language, that is understood by the public you address to so, in this example, the doctors. So I was wondering if you could explain more how you manage that. - So your question is sort of about visual grammar, maybe visual literacy, sort of thinking about the visual representation of things, especially probably my abstract example, and how the doctors or users could understand it.
31:12
Is that what you were asking? Yeah, it's hard. So visual literacy is challenging. And that's something I've had to learn. I mean it's the example Alyssa showed at the beginning with the bar chart. Like you have to learn how to read these statistical plots. And, similarly, like when I do the-- excuse me-- like the network visualization or node link, it does require learning. There is a sort of learning curve to say how can I quickly do that. And it goes back to the context question too.
31:40
And I think a lot of the work I do, and I didn't talk about today, is very much grounded in perception and cognition where I think about literally like what line thicknesses or shapes or colors do I know people will intuitively understand without me explaining it to them. And it's hard to design like that. And it requires lots of user studies and evaluations and throwing designs out and bringing new ones in. But I think, what you're describing,
32:07
it's a huge challenge, but there are sort of techniques for working with it. Yeah? - My question is similar. I have a friend who told me she can't read a map. And I was trying to give her directions, and I didn't know how to get around this obstacle. So I guess my question is there are people that do not want to think abstractly. And all of this visualization is abstract. So could you give an example of why, with the 3D heart arteries and the 2D, why was the 2D easier? - The 2D is easier.
32:39
Let me show you. So the first point is, yes, a lot-- and, one, it's a really interesting point. Different people have different abilities to interpret spatial information and understand abstract representations. It's an actual learning thing and a certain capability we have. So that's always something to bear in mind when designing is to make it as intuitive and easy as possible. And map and cartography design is its own whole area. We take for granted how easy it is to read maps,
33:07
that a poorly designed map is hard to read. So, in the case-- let's see if I can scroll back.
33:17
Let's see. Where was it? They're like here. If you're going from the 3D to the 2D, the reason the 2D, even though it was abstract-- it wasn't anatomically spatial, which is why you need the brushing and linking or context to get back-- a doctor could very easily read left to right, top down, and just look for the darkest red regions. And they instantly could see what it was that they were looking for. They found the diseased regions. For this, I thought through very carefully
33:46
the spatial layout and how the projections-- so that's a pseudocylindrical. It's the equivalent of taking each artery as a cylinder and taking a scalpel and slicing the top and butterflying it out. And that was intuitive for the doctors from their med school training for how they do surgery. And so it took me a lot of interviews and embedding myself in the hospital to figure that out. So there are sort of ecological issues of that nature,
34:10
but then there's things like the color is just perception. Red has a pop out effect. We are sensitive to red. Stop signs and red, and red lights are red. So I knew they didn't even have to think about it. If they saw dark red, they should know it's diseased. So there are sort of multiple layers to the design. I know the other panelists will have similar thoughts and experiences on how to do this design. - So we're just going to take one more question while Arvind plugs in his computer.
34:35
So go ahead, Michelle. - Thanks. - Thank you. I have been going recently to cell biology lectures where people are using new techniques where you can analyze all of the transcripts in single cells and then try to interpret those in terms of an evolution process between one cell going to another. And they end up with two-dimensional images, but there are thousands of parameters. And I don't know how they do that. There's something they call tSNE2. I don't know what that is.
35:06
And I have procrastinated about looking it up. I was wondering if you could give me-- - So I will refer you to Bang Wong, who is sitting in the second row here, who is an expert in this area. And I will give you the short answer, as I'm not an expert on that, is it's hard. I mean it's a massive data problem and so many attributes. Like it's really hard to do, but I'm not-- I don't have a good answer, but, during the break, you should come up to the front. [MUSIC PLAYING] [APPLAUSE]
Here to help you learn
Sign in to keep a private conversation with your course tutor.
Ask me anything about the material…
Sign in to chat