This continues the set of notes I started on the previous post.
It turns out I spent most of Monday hanging out on hallways and chatting with people, so I did not really see much of the sessions. But I will highlight one bit about the BioVis sessions which I thought was pretty great (Robert mentioned this as well on his post). BioVis was run as a traditional symposium, with a set of closely-related papers on a specific area (in this case, biological data visualization). The brilliant idea by the organizers was to start each session with a short presentation by the session chairs. This presentation briefly mentioned all the papers in the session, and put them in context of the wider area. The session chair, having had access to the work and typically more experience in the area than the presenter, can give the wider context. This way the audience can connect dots between the individual papers much more easily. VisWeek should seriously consider doing something like this instead of fast forwards. As much as I enjoyed Gordon’s “Tiny Particles” uke masterpiece, or (shame on me, I forget their name) “The devil came down to Baltimore”, I think I would be more likely to get something out of a short presentation by the session chairs before each set of papers.
I started the day with the Infovis session on evaluation, and with Steve Haroz’s presentation of this year’s best Infovis paper. Haroz and Whitney designed a set of simple experiments that show very clearly how information displays are fundamentally bounded by our limited capacity for attention, and discuss how to optimize visualizations to avoid squandering this limited resource. The paper includes all the relevant details, and you should read it. But what really caught my attention was Haroz’s presentation. I don’t know if VGTC is recording these, but if you have the opportunity to see one of his talks, I urge you to do so: he was clear, persuasive, only explained as much as needed in a talk (because, really, you should read the paper!), and, above all, he used the projector to show, not tell. Really great job. There has been some discussion on how to interpret the results, and Haroz has written a response. How crazy is it that these written discussions can now happen during the course of a conference? We truly live in the future. Also, and I’ll try to put it mildly, dropbox user 14753707: don’t act like a spineless coward, and let us know who you are! This is not how scholarly discourse happens.
The other work I want to highlight of the morning session is Hofmann, Follett, Majumder and Cook’s paper on evaluating visualization designs by appealing to the notion of statistical power. This is a followup and a direct application of the techniques in Graphical Inference for Infovis, which if you’re at all a reader of this blog, you’re probably sick of hearing. But bear with me some more, because this is great stuff. The original paper presented the groundbreaking techniques for turning visual tasks into formal statistical tests. This paper shows how this idea can be used to compare different visualization designs. The basic idea is this: a visualization design is good if it is hard to hide the true data among “impostors” which are created by sampling bogus data with similar distributions to the true dataset. This is almost unbelievably simple, but it turns out to unlock many powerful (and widely studied!) tools from statistical inference theory directly so then can be used in visualization evaluation. As this is an area that in my opinion is in sore lacking of techniques which generalize effectively, this development by Hofmann and co-authors is hugely exciting.
Other cool papers I saw today included (but were not limited to!) the work from Ahmed, Zheng and Mueller in leveraging human computation to design better compositing operators, Wu, Yuan and Ma’s paper on non-independence and interaction modeling for uncertainty displays, and Sedlmair, Meyer and Munzner’s work on collecting best practices for design studies, which are a powerful, popular, and monstrously hard way to do applied visualization work.
Finally, the day ended with a phenomenal party, organized by Noah Illinsky in connection with the Seattle Data Visualization Meetup and graciously sponsored by Tableau Software (if I’m missing anyone else, please let me know!). This was, as Noah put it, an attempt to bring together two groups of people with very similar interests that would otherwise probably not overlap very much. There were five or six talks ranging from things like “why you should do pro bono visualization work and make the world a better place” to the intersection of art and visualization (by Francesca Samsel, who is actually organizing an intriguingly named workshop on Thursday), to our own Robert Kosara talking about story telling in visualization. This was possibly the best party I’ve ever attended at VisWeek, and I would love for something like it to become a tradition.