During VisWeek 2008, I wrote short updates on my website, which I called glimpses (a little pun on tweet). This is an archive of all of those, in reverse chronological order. In my transition to WordPress, I had to get rid of the comments, unfortunately. There were only a small number of those attached to these postings, though.
So that’s it. I will write a bit more about my conclusions from this year’s conference in a regular posting, but I think we have seen a clear shift towards more theory and more cognitive science in InfoVis this year. I expected to see a stronger focus on InfoVis for the masses, after last year’s strong start. There were quite a few papers, but it hasn’t become a major focus yet. That might still happen, though. I hope you enjoyed the live-blog, even if it was spotty at times. This has been an interesting experiment, and one I will definitely try again when there’s a chance. Don’t hold your breath though, there are no conferences coming up in the next few months.
Vis Capstone: David Laidlaw
David Laidlaw gave an interesting, thoughtful capstone talk comparing Vis and InfoVis 2008 to Vis 1998. He did a very extensive study, classifying the different contributions. While he had a lot of interesting observations about why numbers have changed, I felt that his analysis could have gone deeper. He ended with an interesting analogy of a stone wall and its construction, but I think he left quite a few people scratching their heads. Though maybe that’s a good way to end the conference, so people keep thinking about it.
Next Year: Atlantic City, New Schedule
Next year’s conference will be in Atlantic City, NJ. There will be a few changes: the conference will be earlier (Oct 11-16), the schedule changes so VAST and Tutorials will be at the beginning of the week, with InfoVis and Vis perfectly parallel Wednesday to Friday. Vis and InfoVis will also share keynote and capstone talks. This should be interesting, though I’m a bit concerned about how many people will end up attending the workshops and tutorials. Stay tuned for the capstone talk and wrap-up!
Science of Visual Analytics Panel
The last panel of the conference was on the interdisciplinary nature of visualization and visual analytics, in particular with regards to cognitive science. There is clearly a lot of interesting work that we need to connect to (and not just use what’s there, but also inform further research). All this talk about science led to an interesting little side discussion I had with Jarke van Wijk on whether visual analytics should even be a science, or more an artistic or design field. I’m a bit undecided right now. both have their merits.
Tracking Stories As They Unfold
Danyel Fisher from Microsoft presented a paper on how they track narrative events as they unfold. I did not catch the entire presentation, but the examples he showed were quite interesting. The main criterion is to be able to see temporal sequence to tell what is causing what – given tens or hundreds of thousands of news stories, that can be rather challenging.
Losers Don’t Use Visualization
There hasn’t been much activity here today because I’ve been trying to hunt down a particular dataset. This will be turned into a visualization posting here shortly, but it has kept me from paying much attention in the sessions, and I ended up skipping some of them altogether. Savikhin presented a neat little paper at VAST that looked into the effect of visualization (as opposed to simple data tables) in economic decision making – specifically to counter what’s called the winner’s curse (and there’s also a loser’s curse). They found that the visualization was much better than the tables. Not entirely unexpected, but good confirmation and a study that looked quite well-designed.
That’s a lot VASTer!
Now that’s more like it! Anthony Robinson presented some very interesting results from a study he did on analysts working together on paper (with post-its, cut out images, etc.). His results provide a lot of insight into the analytical process, fairly regardless of background. What I found most interesting was that people who had changed or destroyed their own information layout when collaborating later regretted that because they could not go back to their own, complete results. Another presentation by Daniela Oelke on visualizing the precision of text classification and summarization algorithms was somewhat interesting, though their classification problems seemed rather trivial. Somebody then also made the comment that by applying this work to more state-of-the-art algorithms would also help to connect with the computational linguistics community.
Supply Levels Low … Please Replace Cartridge
It’s always interesting to see the break-down of costs of a conference; the Vis/VisWeek organizers started showing those a few years ago. This year again, food accounted for 24% of expenses – a lot more than one might expect. The banquet last night was excellent though, better than what I remember from conferences in recent years. And while I appreciate the need for a bit more frugality this year (after there was apparently a loss carried over from last year’s conference), a bit more coffee and hot water (for tea!) would not cost a whole lot, and would make people happier. Perhaps we could exchange a few of those ginormous muffins (the size of a small country) for another keg of hot water.
Not so VAST!
The first VAST session this year was less than exciting. The presented papers mostly did analysis using machine learning or statistics, but the visual parts were rather poor and seemed to have been tacked on at the end to produce an image. One paper in particular, Garg et al’s Model-Driven Visual Analytics, looked like a simplified version of Xiao et al’s paper at VAST 2006 that did not require models to be built by hand, but rather had a compelling visual interface for that.
It’s almost like being there – a year ago
vgtc (the visualization and graphics technical committee – all lowercase) has a website where you can look at the slides from the talks of last year’s conference (and also VR and 3DUI). They’re recording the talks this year and collecting slides, and those should be available on that website in a few weeks. In the future, they might even do live streaming (and eventually, according to Amitabh Varshney, have the videos there before the actual talks).
Vis Keynote: Margaret Livingstone
The Vis keynote was wonderfully interesting, entertaining, and a perfect fit for this community. Margaret Livingstone talked about basic perception and showed how artists have (often inadvertently) figured out many things we are only now discovering to make more effective images. Her book should also be interesting. Carlos has minute-by-minute coverage of everything that happened in that session.
The Conference Hotel
I’m not generally very fond of the conference hotels, because they all tend to look the same. The Columbus Hyatt Regency is not a huge exception, though it is rather attractive from the outside. A pleasant surprise was their ballroom, where the banquet was held tonight: instead of the usual hideous, baroque Versailles-wannabe with mirrors and ornaments, they went for a modern layout and look that felt like it actually belonged into this century. They couldn’t do without big-ass chandeliers of course, but I’ll forgive them that. What’s harder to forgive is the way the floor shakes and the explosion noises coming from the A/C in the room used for the tutorial today. I appreciate the technical difficulties in building an A/C for such a large building, but this is beyond annoying.
Day 3 Wrap-Up
After three days, I average about eight postings per day. That’s not so bad, and actually more than I had expected. There is a bit more delay than I would like, but I’m also here to pay attention, talk to people, etc., so the blog sometimes has to wait. The conference has been great so far, we’ve been very successful and I’ve gotten a lot of excellent feedback about our work. I’ve also seen a few very interesting presentations and had great discussions about topics like the theory of visualization, the importance of design, etc. Well worth coming here.
Spatially Ordered Treemaps
Despite their name, treemaps are not maps. Jo Wood and Jason Dykes tried to merge treemaps with maps by preserving the spatial layout and other features (like the relative position and bearing) when laying out georeferenced data in treemaps. My feeling is that they introduced so much additional information, and distorted the map so much, that a system with two views and brushing would have been easier to read and more useful. Distorting the main visual cue on a map – space – also seems to be a bad idea to me in general, whether it’s a treemap or some other distortion.
User-Generated Graph Layouts
Frank van Ham presented work today he had done with Bernice Rogowitz on letting users lay out a network (called a graph in InfoVis/computer science – I had an interesting discussion about that today with Stephen Few) on IBM’s Many Eyes website. Their goal was to find out how the common wisdom in graph drawing meshed with what users would design. But the most interesting thing they found was that semantics matter! They used male and female first names as labels for the network, and they found that their users interpreted these in different ways, rather than treating them as random labels. Who would have thought? But seriously, that’s an important lesson for anyone designing user studies.
Best Student Poster at VAST for Alex Godwin
Alex Godwin got the Best Student Poster Award at VAST for his Interactive Poster: Visual Data Mining of Unevenly-Spaced Event Sequences (Alex Godwin, Remco Chang, Robert Kosara, William Ribarsky). Well done! More updates when I can find a power outlet.
Functional Streaming Tutorial
I decided to skip the combined InfoVis capstone/VAST keynote to attend the Functional Streaming tutorial (materials available from the presenter group’s publications page). Functional programming is really fascinating, and it lends itself well to stream processing. The tutorial was quite dense, but I will certainly have to look into this more. While they concentrated on SciVis (marching squares and cubes, in particular), I think a lot of these concepts would also work well in InfoVis.
Effectiveness of Animation for Trend Visualization
George Robertson presented work his group has done to investigate the effect of animation on the presentation and analysis of trends. This was in response to the highly effective use of animation by Hans Rosling in his TED talks – despite the general wariness of visualization and cognition researchers towards animation. They found that animation was significantly more effective than traces or small multiples for presentation, but worse than the other two for analysis. That is a very interesting result, because so far people have not really looked at visualization for presentation separately from analysis, and it shows that we can make presentations of data more effective (in addition to being more engaging) by using animation.
Bumping into Projectors
I have never seen so many people bump into projectors. It seems that in every single session, somebody gets up and bumps into a projector, so that a part of the projection is on the border of the screen, or even on the wall. These things aren’t really hard to see, either. I don’t get this.
Cloning Nodes for Reduced Clutter in Graphs
Natalie Henry gave a very entertaining and interesting presentation of a study she did to test different ways of showing nodes that connect communities in a large network. The problem is that these nodes can produce the impression that the two communities are connected “in depth,” even if there is only one node connecting the two. She tried different strategies, some of which were mostly confusing, but their CloneLink approach (which is a nice example of perceptual layering) turned out to work quite well.
Grand Challenges Panel
There is not enough room in this microblog posting for a complete description of that panel – but it was interesting. Georges Grinstein kicked it off by making some good remarks about the nature of Grand Challenges, falsifiable theories, and that you need an idea of the direction in addition to the question. Tamara Munzner gave an excellent presentation (available on her talks page) about outward vs. inward challenges, and presented a few good key criteria (described well on her slides). Daniel Keim criticized her position by saying that she was talking about writing papers not unifying visualization theory, and he’s right – still, it’s a great start. Keim also went on to talk about how visualization can save the world, and in doing that he picked up Munzner’s idea of the “infovis-hard” problem of total political transparency (again, see her slides). Then George Grinstein came back to talk about his ideas for how we can build such a science of visualization, and that’s where he lost a good part of the audience. His main idea is to “measure everything,” and to use benchmark datasets to drive the development of theory. I don’t entirely disagree, but I think some of his comparisons with other fields (like KDD and computer vision) were not 100% applicable. Still, a great panel, and I believe Grinstein can at least make more progress than we have in the last ten years put together. To (roughly) quote him, “we’ve got enough facts, it’s time to build a science!”
Shaping of Information
I don’t know if I’ve mentioned it, but we had a paper in the Design session today. Caroline did an excellent presentation, and we got six or seven questions (plus at least the same number again of people who walked up to us afterwards to ask questions and make suggestions). While the questions were largely focused on things we could have done differently (especially Ben Shneiderman’s), they were all very constructive and certainly very valid. This is only a first step, and I think we’re on to something here (and I’m not the only one who thinks that). So this could have hardly gone any better: a lot of people are going to think about this, and hopefully this will spark a bunch of similar experiments. We can certainly use more work on representation, metaphors, etc.
An interesting presentation was Cerebral by Aaron Barsky, Tamara Munzner, and others. They visualize protein interactions inside the cell, and do that over time. The results are being used by biologists to test drugs, and their images have also been used in biology papers, which is a great proof that the program is useful. It’s also available as open source (from the link above). Good stuff.
Stacked Graphs, Vispedia, Word Tree @ Scheidegger
Carlos Scheidegger has blogged about Lee Byron’s excellent Stacked Graphs talk, and also covers two papers from the text visualization session (Vispedia and Word Tree), which I missed while I was at the Design, Vision, and Visualization Workshop. He also has a different take on the keynote. And he cut his hair, so he doesn’t look like the picture on his website at all. Pretty sneaky.
More coming tonight …
Too many things are happening right now, I’ll catch up with more posting tonight. The Design session was very interesting, and now it’s the Grand Challenges Panel, which I’ll certainly write about too. So hang in there!
The workshop on Design, Vision, and Visualization in the morning (and through the lunch break) was very interesting. We talked about design decisions, tweaks, and redesigns of visualizations. The presentations were very strongly geared towards applications, which was a great addition to the mostly academic work shown in the papers sessions. Unfortunately, there was not nearly enough time for discussions, though we did collect a good list of issues to look into further. And, there is obviously quite a bit of interest in design, but we just haven’t made the connection too well so far.
Jake Kolojejchick of Viz (formerly MAYA Viz) presented some good points, but most of his examples centered around maps. He also spent a lot of time talking about their Command Post of the Future work, which isn’t exactly new (Steve Roth presented that in his capstone at InfoVis 2004). Their work is undoubtedly very useful and interesting, but it’s also clear that the main thing military users in the battlefield need are maps – the affordances these need to provide are kind of obvious. A more balanced view with more InfoVis-y topics would have been more interesting. Collaboration is much easier on maps, because you have a ground truth (no pun intended) that you work on, which is quite different from most other visualizations.
Day 1 Wrap-Up
This has been an interesting day. The conference started with some really good papers (and a few so-so ones), and the liveblogging went quite well. It’s quite a bit of work to do this, because I’m writing these things while I’m listening (sometimes to the talk I’m writing about, sometimes to the next one), but it also forces me to put my thoughts into more or less coherent sentences quickly. There were ten postings today (not counting this one), which is a bit more than I expected. I’m still trying to figure out what the best size is for them, but around 100-120 words seems a good length. Maybe a bit long for a true “microblog,” but that’s why I like making my own rules ;)
The best poster this year is by Maria Velez et al, and addresses visual and visualization literacy. Their study looked at the problems people have with visualization to understand how they can transfer knowledge about one visualization to faster understanding of another. This is a very important topic, and it’s great to see some work in this area; I hope this will be a paper next year.
InfoVis Contest 2.0
When I complained some time ago about the sad state of the InfoVis Contest, I wasn’t kidding. We just presented the single submission we got (Patrick Gage Kelley and Daniel R. Rashid), who got an Honorable Mention. I’m stepping down as a co-chair after my two years, and the team of Petra Isenberg (who is new), Bongshin Lee, and Jing Yang are changing the contest to an exhibit. They will no longer publish a dataset, but rather ask for submissions of interesting insights gained with visualization tools. They want to collect a portfolio of uses of InfoVis, which should be interesting.
Who Votes for Donut Charts?
John Peltier would be jumping up and down if he were here, seeing Geoffrey Draper present his work on visualizing exit poll data. He’s using what is essentially a donut chart for the data. The interaction is not bad, but the visualization is really close to useless. You can’t see different subsets at the same time, and the donut makes comparing between different categories difficult. There are much better ways to show this kind of data, like Parallel Sets. Their goal was for this to be useful for non-experts, which it undoubtedly was – but I wonder if a better visualization might get close in that regard and still be a better visualization.
Those Who Forget History …
Heer’s work on history management in Tableau is interesting. There seems to be a lot of interest this year in recording sessions and using that information for various purposes. Good extension of long-standing work in HCI – I wonder when this will make it into the program.
Distributed Cognition for InfoVis Theory
Criticism of the sad state of theory in InfoVis. Yes! Zhicheng Liu presented work he has done with Nancy Nersessian, John Stasko. His take on internal and external knowledge and the way they can be applied in InfoVis is a great basis for further work. He also quoted from our paper (yes, another link to our stuff, but what can I do?). Stasko and his students have been doing some excellent foundational work in the last years, it’s great to see that.
Interaction with Probes
I’m pointing to UNCC work again, but I’m liking this more and more. Tom Butkiewicz presented his work on using probes to query simulations on maps (PDF). This is a good approach to multiple views and allows a lot more useful insight by seeing and comparing local effects. Well presented, too!
An Interesting Take on Scatterplots
The Best Paper by Elmqvist, Dragicevic, and Fekete has an interesting take on using 3D transitions for navigating scatterplot matrices. The target audience is also people not too familiar with visualization, though I think this would still overwhelm them. The actual navigation between cells isn’t improved, and Georges Grinstein made a valid remark about the cognitive load when switching between 2D and 3D. But it’s good to see more animation for navigation, Robertson et al.’s paper later will also be interesting.
Yay! Best Paper Honorable Mention!
Papers Fast Forward Starts with a Bang
Wow, the Papers Fast Forward starts with a bang! Extremely well done “movie opening” for the best paper, certainly the most elaborate paper preview I have ever seen. Jeff Heer also did a great job with his minimalist list of quotations (and no talking). Caroline also did a great cliffhanger-style preview of our paper. Looks like it’s going to be an interesting few days.
And so it begins …
The annual gathering of the InfoVis family has begun. Lots of familiar faces. One change from the previous years: there is no conference bag. I actually think that that’s a good thing, because I tend to not use them (they’re usually small and not the prettiest or sturdiest) and just throw them away after a year – when I get the next one.