Discussion of Visualization Designs

From CS 294-10 Visualization Sp10

Jump to: navigation, search

Lecture on Feb 10, 2010




Same readings as Interaction lecture. Finish reading whatever you missed from last time. You are still responsible for a comment for today even though the readings are the same.

  • Visual information seeking: Tight coupling of dynamic query filters with starfield displays, Ahlberg & Shneiderman. (html)
  • Visual exploration of time-series data, Hochheiser & Schneiderman. (html) (pdf)
  • Postmortem of an example, Bertin (pdf)
  • The visual design and control of the trellis display. Becker, Cleveland and Shyu. (ps)

Optional Readings

  • Table lens, Rao and Card, (acm)
  • Human guided search: Survey and recent results, Klau, Lesch, Marks & Mitzenmacher. (pdf)
  • Design and evaluation of incremental data structures and algorithms for dynamic query interfaces. Tanin, Beigel & Schneiderman (citeseer)


Chetan Nandakumar- Feb 09, 2010 09:13:50 pm

When visualizations get to be interactive, a big design issue becomes human memory -- the fidelity of memory representations bears a huge weight on how to effectively manipulate the information presented. From previous readings, it seems that visual memory can be quite poor. First off, what are these perceptual constraints? How do these constraints reflect in the design of effective interactive visualizations?

Danielle Christianson - Feb 10, 2010 09:08:11 am

While I think automated and interactive visualization tools are incredibly useful, I'm wondering what kind of traps they set up. I.e., A user could become too reliant on the software for data analyze and miss visualizations that are not prescribed in the software. This is a typical concern with new technologies and obviously is does not mean the demise of critical thinking, but I'm wondering if there is something that can be included in automated, interactive visualizations that can encourage the user to think more broadly than the software's limits.

Jeffrey Patzer - Feb 10, 2010 10:38:48 am

I am surprised that we do not see more use of time boxes by scientific researchers in labs or other domains. Considering the power to cull through one's data, and pick out trends or instances, it begs the question why microsoft excel or another popular software package hasn't implement one of these dynamic interaction tools. Seeing how tedious data entry is, it seems that these sort of tools could leverage the standard set by such popular programs to create more powerful displays than they are currently doing. After reading these various articles that are so many years old, that any of these systems have not been implemented in more mainstream consumer software (especially considering the high usage of time-series data).

Stephen Chu - Feb 10, 2010 11:39:15 am

I find Google Maps to be a very good example of a tool that offers effective visual information seeking capabilities. Ahlberg and Schneiderman outline the following keys for direct manipulation: 1) visual representation of the world of action including both the objects and actions (map of world, zoom in/out, street view, terrain, etc.) 2) rapid, incremental and reversible actions (easy to zoom in/out, reverse direction of street view, change view type) 3) selection by pointing (not typing) (yes, gmaps uses more pointing, but specific addresses must be typed) 4) immediate and continuous display of results (if your internet connection is fast enough) 5) dynamic query filters: query parameters are rapidly adjusted with sliders, buttons, etc. (zoom in and out can be done through a slider, street view can be used with click of a button) 6) starfield display: result sets are continuously available and support viewing of hundreds or thousands of items (roads, highways, Google Buzzes on Android!!) 7) tight coupling: query components are interrelated in ways that preserve display invariants and support progressive refinement. (Output can always be manipulated as input)

Paul Ivanov - Feb 10, 2010 12:37:14 pm

Hochheiser and Shneiderman: great article. One other great application for time boxes, outside of exploring data, is data filtering (though I suppose filtering is a type of query process, too). I've seen exactly this principle applied in electrophisiology software. An electrode placed inserted into the brain to record activity. The problem is to isolate the activity of a single neuron, because the electrode is like an antenna, and 'sees' the activity of all the neurons around as a change in the voltage due to a neural spike. Fortunately, the *shape* of the spike, at least according to most people in the field, is slightly different for different neurons. Neurons further away have different shape to the onset relative to the peak, lower peak, etc. The experimenter isolates the neuron of interest using time boxes to filter only the data of interest as the recording proceeds. I'm not sure whether this type of thing was used before the article, but it was a really neat approach to see used in data collection, not just data analysis.

Aaron Hong - Feb 10, 2010 12:33:43 pm

What is interesting about the chapter in Bertin's Graphics and Graphics Information-Processing is that often times this information processing does not involve the machine. "Thus the most important stages--choice of questiosn and data, interpretation and decision-making--can never be automated. There is no artificial 'intelligence.'" Much of what it takes to solve a problem using visualization takes what he calls "imagination," which the machine is not capable of. That is why interaction is important and why the process of creating the visualization itself is solving a problem, not just giving information for people to use to solve a problem. Basically, what I'm saying is that once a visualization is done it should have solved the problem already. "A graphic designed for communication highlights the main points of the results of the investigation."

Arpad Kovacs - Feb 10, 2010 12:52:30 pm

Time boxes seem like an excellent idea for filtering quantitative-quantitative or ordinal-quantitative data. However, these situations are not very common in real life; for example how often would an investor search for a set of stocks that closed at a particular value on a particular day? Instead, users generally want to compare a few distinct values and look for overall trends. I think this is the same reason why the alphaslider has not come into wider use; these widgets are more suitable for selecting a range of values rather than single discrete variables, which the ubiquitous toggle checkboxes and drop-down lists are better suited for.

Zev Winkelman - Feb 11, 2010 01:08:00 pm

I was very impressed with the sample of my fellow classmates' work today.

It was particularly enlightening to see how people who had chosen similar designs approached the encoding of information in different ways.

Looking forward to learning more from everyone throughout the rest of the semester.

Jon Barron - Feb 11, 2010 12:52:43 pm

Human Guided Search: I had really high hopes for this paper. Basically, they allow users to visualize the current state in combinatorial optimization problems, and the user can assign "mobilities" to different components of the search problem, which determine how much (if at all) those components shift around. This hybrid optimization works better than either individual search method. Unfortunately, this paper doesn't have much to say about either visualization of optimization, two topics I do care about, and instead focuses on the details of the particular discrete optimization problems they used, which I don't really care for. I would have appreciated some study of how the choice of visualization affects optimization (presumably, showing the user the bytecode of the optimization problem would not help much) but there is nothing. They don't even seem to draw connections between and lessons from the visualization techniques they use in the different problems, and instead are just content to make ad-hoc visualizations for each problem. I would have also liked some investigation into how users can assist in optimization over continuous or hybrid systems, which are things I (and most people?) care more about.

Overall though, I like the idea. This paper reminds me of "Advanced Chess", in a good way.

Jonyen - Feb 12, 2010 09:06:42 am

I think the "Visual Exploration of Time-Series Data" webpage was kind of hard to make anything out of it because it's all static images. Compare with more dynamic images such as http://cohort11.americanobserver.net/latoyaegwuekwe/multimediafinal.html. With time-related data, I feel like it is better understood when you show some animation or movement so that it correlates with the actual passing of time as you're watching the visualization. But I guess it doesn't always work out in every case because sometimes you may want to juxtapose data from different times to better compare the data.

Akshay Kannan - Feb 15, 2010 12:03:47 am

Encoding time has traditionally been done on the x-axis, as shown in most visualization software we explored. However, with interactivity and animation at our disposal, there may be some alternative (and possibly more informative) ways of encoding time. Most obvious, would be simply encoding time with time, where the designer creates an animated visualization which changes over time to reflect changes in a given period (as in a time lapse video). Also, another interesting possibility for showing time would be utilizing depth, (as in Apple's time machine: http://www.winsupersite.com/images/showcase/leopard_17.jpg). While the x-axis representation of time is most familiar to viewers, there are plenty of other interesting representations to be explored.

Prahalika Reddy - Feb 15, 2010 01:34:39 am

The Bertin reading seems like it could be useful when we're woking on Assignment 3. The process followed by the hotel manager's assistant is essentially what we will need to do when finding our data and trying to figure out what question we want to answer. I especially like the idea that a graphic isn't just "drawn once and for all", but that is it "constructed" using the image's mobility. I think that's one of the key ideas to finding trends; to be able to rearrange data easily to see it with a different view can be really helpful when looking for a certain trend.

I really liked the interactive tools we saw in lecture, especially the Zipdecode. The concept of Zipdecode is a simple one which could be done almost as effectively with the user entering text and the system returning a location in text, but the way it's done interactively is really very nice. It's good to be able to see the system narrowing in on a certain location as the user types in digits. Also interesting was the NameVoyager, though it seems kind of complicated to me. At first glance, it's not very clear how all the data is arranged to make the visual. It is however a very interesting concept and is still visually effective.

Ebby Amirebrahimi - Feb 15, 2010 12:45:52 pm

I wanted to take a look and analyze the demonstrations:

Homefinder was a good historical piece almost of how interactive visuals can be made useful. It demonstrates the usefulness of dynamic queries.

The Cellphone finder is perhaps an example of how an interactive visualization can get too complex. I felt like there were too many options concentrated at the top of the screen and it was not pleasing to look at. I think it would have been better if they spread all these options over a series of pages.

I thought Fry's Zipcode finder was absolutely fantastic. The application feels alive and responsive. It's fun to see the map zoom when typing numbers, and the map itself is visually appealing. I think this is a great example of how dynamic queries can improve user experience.

Finally, table lens appears to be a very powerful BI tool. In this example: http://www.sapdesignguild.org/community/book_people/visualization/images/table_lens.gif It's clear that this tool could potentially reveal some very important correlations in a data set.

Priyanka Reddy - Feb 16, 2010 09:34:18 pm

The optional paper on human guided search proposes a great idea: allow humans to provide input to optimization systems to provide the best solutions. These systems are known as interactive or human-in-the-loop optimization. I think this happens to be a very practical solution - it allows computers to do what they do faster and humans to jump in on what they do faster. There are many problems that are hard to solve using a computer alone, which can be improved drastically with human input (ie. differentiating between "apple" the fruit and "apple" the company in a search engine). And, now with the increase of smartphones, which have much lower computing power, tasks that you could do quickly on your desktop become much slower on the phones. In this case, allowing humans to input some contextual information can really save battery life. Granted, a fully automated system is usually preferable (assuming the user trusts the system), but remembering and using this intermediate interactive system as an option can be beneficial.

The paper also mentions the concept of trust and whether or not the user will trust the output of a system without knowing what the process was. In the US, I think more people are able to trust their computers to do more and more complicated tasks, so we don't see the issue of trust as much. However, in developing countries, where many people use cell phones, the issue of trust still exists. Allowing users to really understand the process taking place behind the scenes is a great potential solution to this problem.

Boaz Avital - Feb 18, 2010 03:04:05 am

To me, the trellis display seems to be a sort of formalization of small multiples. I especially liked the suggestions for ordering of data that doesn't otherwise have an obvious display order.

The reading by Tanin, Beigel & Schneiderman was a pretty innovative and interesting system. Fast efficient querying is already a problem with large datasets, so to extend into incremental querying and keep the system fast and responsive is impressive. The conclusion of the paper mentioned that the next thing they wanted to do was increase the speed by a whole order of magnitude, and I am interested to know if they succeeded.

[add comment]
Personal tools