Here are some of my comments:
Comment on Jill's Blog
Comment on Eric's Blog
Comment on Patrick's Blog
Comment on Jesus's Blog
Comment on Jesus's Blog
Comment on Justin's Blog
Comment on Kerry's Blog
Comment on David's Blog
Comment on Gus's Blog
Comment on Justin's Blog
Comment on Bill's Blog
Comment on Jacob's Blog
Comment on Jacob's Blog
Comment on Ross's Blog
Comment on Jill's Blog
Sunday, March 21, 2010
Baseball Cards [✯IUI '09]
Rich interfaces for reading news on the web by Earl Wagner, Jiahui Liu, Larry Birnbaum, and Kenneth D. Forbus; Northwestern University.
Summary:
The authors of this paper created a system, called Brussell, that assisted in reading news articles by providing additional information and summary background information from within the article and other related articles. They started by defining several templates for article types (kidnapping, acquisition, etc) for testing. For each of these templates there is a set of a information such as previous events. This information can be used to provide additional details to the users or to give a simple overview. (Like the information in the box on the right side of Wikipedia examples.) The information is gathered from previous articles. The articles are cross-referenced to remove incorrect information. When a current article is processed relevant information is linked. The benefit of all this is that people can find background information for current articles. To test the system they first used a database of old articles to build a knowledge base for Brussell. Then they looked at 100 articles to see how many "references" were identified by Brussell. The mean was 4.1 per article and the median was 3 per article.
Discussion:
This article was pretty dry. The idea definitely could be useful, but it seems like a lot of effort for something so small. It would really be idea to use a central database for everyone and allow authors of articles to add information. This way it could be made even better than the technology allows and if technology becomes better than human tagging then people don't have to do any work. (Basically, it would just be nice if now the authors could confirm information and help train it.) I liked the idea they had of right clicking on things and getting suggestions such as "what happened in ____ ."
Summary:
The authors of this paper created a system, called Brussell, that assisted in reading news articles by providing additional information and summary background information from within the article and other related articles. They started by defining several templates for article types (kidnapping, acquisition, etc) for testing. For each of these templates there is a set of a information such as previous events. This information can be used to provide additional details to the users or to give a simple overview. (Like the information in the box on the right side of Wikipedia examples.) The information is gathered from previous articles. The articles are cross-referenced to remove incorrect information. When a current article is processed relevant information is linked. The benefit of all this is that people can find background information for current articles. To test the system they first used a database of old articles to build a knowledge base for Brussell. Then they looked at 100 articles to see how many "references" were identified by Brussell. The mean was 4.1 per article and the median was 3 per article.
Discussion:
This article was pretty dry. The idea definitely could be useful, but it seems like a lot of effort for something so small. It would really be idea to use a central database for everyone and allow authors of articles to add information. This way it could be made even better than the technology allows and if technology becomes better than human tagging then people don't have to do any work. (Basically, it would just be nice if now the authors could confirm information and help train it.) I liked the idea they had of right clicking on things and getting suggestions such as "what happened in ____ ."
Friday, March 19, 2010
A Brief History of History [✯UIST '08]
An application-independent system for visualizing user operation history by Toshio Nakamura and Takeo Igarashi.
Summary:
In this paper, the authors create a generic system for visually annotating operation history. In simpler terms, they used images for undo/redo type history and put notes on them to help improve utility. The paper discusses the limitations of prior works such as a visual history being tied to a single program. They create a system for annotating visual histories by adding symbols that indicate what the user did at that stage. The system was independent of any application and used a combination of mouse/keyboard monitoring and screenshots to create the annotated visual history. The main annotations were arrows to represent dragging and stars to represent clicking. They performed a user study to determine if the annotations would improve user performance over a visual history without annotations. The users were half expert (graduate computer science students) and half novice (non-cs undergrads that "rarely used" computers). They had three different types of GUIs that they gave the users to test the history with. The three GUI's (seen in the second image) were a 3D drawing program (a), an icon manipulation system (b) and a GUI widget system (c). The users were given tasks to find in a visual history and their performance was recorded. The results were that both novice and expert users had overall improved speed and reduced error rate with the annotated histories. The best performance improvements were with the 3D drawing program (a). In this case errors and time were cut to a third for both novice and expert users when using annotations. The other cases did not do as well. In the icon GUI (b) the time was better with annotations but error rate was actually higher. The researches postulated this may have been due to the cryptic nature of the stars to the novice users (since there were different representations for single, double and right click) and the fact that some icons were obscured by the annotations. For the GUI widgets (c) there was no significant improvement in speed but some improvements in error rate. Overall they determined that the system has some definite applications and can be applied without knowledge of the task being performed.
Discussion:
First, let me say I don't recommend the video. The paper makes many good points and helps progress something that is complex for users in a fairly straightforward way. They didn't have to make plugins or use special cameras they just wrote a small program that acts as a middle man. Histories can be difficult to understand when not visualized. I can look through my Photoshop history and see my last 20 operations were selections, but that doesn't help me: I want to know what I selected. They expanded upon basic visualization, however, filling in gaps where it might not be clear how the state changed from just two snapshots. It makes sense that it would be most useful in 3D programs because that is where such visual histories are most cryptic. If you have a shot of the front of an object and then the back you need something to tell you that the view just changed. A simple arrow (as used) does the trick because, after all, it mimics the natural dragging movement the user made to perform the state change. The downside of the visual history is it is large and can be unmanageable for large numbers of operations, but these are limitations the researchers have recognized and hope to address in the future.
Summary:
In this paper, the authors create a generic system for visually annotating operation history. In simpler terms, they used images for undo/redo type history and put notes on them to help improve utility. The paper discusses the limitations of prior works such as a visual history being tied to a single program. They create a system for annotating visual histories by adding symbols that indicate what the user did at that stage. The system was independent of any application and used a combination of mouse/keyboard monitoring and screenshots to create the annotated visual history. The main annotations were arrows to represent dragging and stars to represent clicking. They performed a user study to determine if the annotations would improve user performance over a visual history without annotations. The users were half expert (graduate computer science students) and half novice (non-cs undergrads that "rarely used" computers). They had three different types of GUIs that they gave the users to test the history with. The three GUI's (seen in the second image) were a 3D drawing program (a), an icon manipulation system (b) and a GUI widget system (c). The users were given tasks to find in a visual history and their performance was recorded. The results were that both novice and expert users had overall improved speed and reduced error rate with the annotated histories. The best performance improvements were with the 3D drawing program (a). In this case errors and time were cut to a third for both novice and expert users when using annotations. The other cases did not do as well. In the icon GUI (b) the time was better with annotations but error rate was actually higher. The researches postulated this may have been due to the cryptic nature of the stars to the novice users (since there were different representations for single, double and right click) and the fact that some icons were obscured by the annotations. For the GUI widgets (c) there was no significant improvement in speed but some improvements in error rate. Overall they determined that the system has some definite applications and can be applied without knowledge of the task being performed.
Discussion:
First, let me say I don't recommend the video. The paper makes many good points and helps progress something that is complex for users in a fairly straightforward way. They didn't have to make plugins or use special cameras they just wrote a small program that acts as a middle man. Histories can be difficult to understand when not visualized. I can look through my Photoshop history and see my last 20 operations were selections, but that doesn't help me: I want to know what I selected. They expanded upon basic visualization, however, filling in gaps where it might not be clear how the state changed from just two snapshots. It makes sense that it would be most useful in 3D programs because that is where such visual histories are most cryptic. If you have a shot of the front of an object and then the back you need something to tell you that the view just changed. A simple arrow (as used) does the trick because, after all, it mimics the natural dragging movement the user made to perform the state change. The downside of the visual history is it is large and can be unmanageable for large numbers of operations, but these are limitations the researchers have recognized and hope to address in the future.
Advanced Meta Scripting or The Ultimate Computer [IUI '09]
From geek to sleek: integrating task learning tools to support end users in real-world applications by Aaron Spaulding, SRI International, et al.
Summary:
This paper discussed how the authors created a system to assist military personnel in automating tasks by creating procedures. Their system, called ITL (for Integrated Task Learning) combined the utility of several automation systems to provide a common system for creating learned procedures. Learning procedures involves having users perform those procedures while the system observes and then the system attempts to programmatically repeat them. The system makes loops, assigns variables and determines choices to be made based on the examples given. The user can then modify what the computer has created for better and more correct functionality. For their contributions they simplified the textual display of the steps to increase human readability, added a graphical interface for adding new steps and restrictions, and enabled users to copy steps between procedures. None of these were trivial and they involved both interaction with the users to gain an understanding of what would be useful as well as intelligent processing of the procedures. For instance, to copy the steps from one procedure to another required that the program attempt to map variables from one context to another.
Discussion:
This paper was okay. I don't know how useful this is because even as simple as it is there are still few people that would find such a system useful, especially off the clock. A problem faced by simple scripts and automations in general is that they are limited in what they can do by the dictionary, the programming paradigms and the clarity of what they do. In other words the more you simplify the less control users have. This means it may work really well for some things but for others users will hit a dead end for the automation. However, this system does have distinct advantages. It may seem more appropriate to have "technical people" create scripts programmatically but given the simplicity of such mechanism (in terms of programming abilities) it is all too easy for those who don't know what they're doing to cause problems. With a procedures system in which the actual execution is done by a separate program, security and stability issues can be addressed and even controlled on a global level. In addition, since the creation and maintenance of the procedures is placed in the hands of the users there is no room for miscommunication between the user and developer. This means the only limits are in the tools, which is what these researchers aim to improve.
Summary:
This paper discussed how the authors created a system to assist military personnel in automating tasks by creating procedures. Their system, called ITL (for Integrated Task Learning) combined the utility of several automation systems to provide a common system for creating learned procedures. Learning procedures involves having users perform those procedures while the system observes and then the system attempts to programmatically repeat them. The system makes loops, assigns variables and determines choices to be made based on the examples given. The user can then modify what the computer has created for better and more correct functionality. For their contributions they simplified the textual display of the steps to increase human readability, added a graphical interface for adding new steps and restrictions, and enabled users to copy steps between procedures. None of these were trivial and they involved both interaction with the users to gain an understanding of what would be useful as well as intelligent processing of the procedures. For instance, to copy the steps from one procedure to another required that the program attempt to map variables from one context to another.
Discussion:
This paper was okay. I don't know how useful this is because even as simple as it is there are still few people that would find such a system useful, especially off the clock. A problem faced by simple scripts and automations in general is that they are limited in what they can do by the dictionary, the programming paradigms and the clarity of what they do. In other words the more you simplify the less control users have. This means it may work really well for some things but for others users will hit a dead end for the automation. However, this system does have distinct advantages. It may seem more appropriate to have "technical people" create scripts programmatically but given the simplicity of such mechanism (in terms of programming abilities) it is all too easy for those who don't know what they're doing to cause problems. With a procedures system in which the actual execution is done by a separate program, security and stability issues can be addressed and even controlled on a global level. In addition, since the creation and maintenance of the procedures is placed in the hands of the users there is no room for miscommunication between the user and developer. This means the only limits are in the tools, which is what these researchers aim to improve.
09:09 on 09/09/1999 [IUI '09]
Data-driven exploration of musical chord sequences by Eric Nichols, Indiana University; Dan Morris and Sumit Basu, Microsoft Research.
Summary:
In this paper the authors created a system whereby new musical cords could be created based on combinations between chord characteristics of preset groups. They had a 2D "polygon slider" (think radar chart) that controlled the mix of features between different "genres" to create an overall chord sequence. They created these genre based axis in four different ways: human tagged genres, genres with AbsDiff Clustering, Principal Components Analysis (PCA) and random groupings with AbsDiff Clustering. The user study had 10 users use each of these four methods and move the points in the polygon slider to create chord sequences. Everything was blind in this study and the axis were not labeled by the genres but rather as "A", "B", etc. They found that all participants preferred the computer assisted groupings to the label based genres alone. These allowed for greater creativity and better "exploration" by participants. The PCA was less favored for being too sensitive to small changes. Study participants noted they would like more fine controls such as locking certain parts of a chord or limiting the affect of continued changes.
Discussion:
This was an interesting way of looking at music and an interesting use of their evaluations. It seems like this would also be a fun tool for those without a musical background to allow them to create simple chord based rhythms. I have used programs to create simple compositions for use as background music in videos, but that is mostly just combining loops. Since I don't have a knowledge of chords I am stuck with adding simple melodies if I want anything more, even if I know what I want. Still, this use, although interesting, seems limited in utility. The ability of a computer to understand properties of the music and aid in composition is what could really be interesting.
Summary:
In this paper the authors created a system whereby new musical cords could be created based on combinations between chord characteristics of preset groups. They had a 2D "polygon slider" (think radar chart) that controlled the mix of features between different "genres" to create an overall chord sequence. They created these genre based axis in four different ways: human tagged genres, genres with AbsDiff Clustering, Principal Components Analysis (PCA) and random groupings with AbsDiff Clustering. The user study had 10 users use each of these four methods and move the points in the polygon slider to create chord sequences. Everything was blind in this study and the axis were not labeled by the genres but rather as "A", "B", etc. They found that all participants preferred the computer assisted groupings to the label based genres alone. These allowed for greater creativity and better "exploration" by participants. The PCA was less favored for being too sensitive to small changes. Study participants noted they would like more fine controls such as locking certain parts of a chord or limiting the affect of continued changes.
Discussion:
This was an interesting way of looking at music and an interesting use of their evaluations. It seems like this would also be a fun tool for those without a musical background to allow them to create simple chord based rhythms. I have used programs to create simple compositions for use as background music in videos, but that is mostly just combining loops. Since I don't have a knowledge of chords I am stuck with adding simple melodies if I want anything more, even if I know what I want. Still, this use, although interesting, seems limited in utility. The ability of a computer to understand properties of the music and aid in composition is what could really be interesting.
Saturday, March 13, 2010
I'm A Frayed Knot [IUI '09]
Pulling strings from a tangle: visualizing a personal music listening history by Dominikus Baur and Andreas Butz, University of Munich.
Summary:
This paper proposes a means of using a visual history to allow a media program to create playlists based on the previous listening sessions of a user. They begin with the prior works, notably Last.fm and Pandora, and move on to the methods they used to allow users to visualize thier listening habits and then create playlists. The first method they use to visual listening history is a tangle. To create a tangle they connect all songs that are listened in succession to each other (directionally). This means that if you listen to "Save A Horse (Ride A Cowboy)" and then "The Rockafeller Skank" and then "The Spirit Of Radio" then they are connected in that order. If you then listen to "The Everglow" and then "The Rockafeller Skank" then the "The Everglow" is connected to a new visual representation of "The Rockafeller Skank." Tangles form in the chain of songs when the different portions are brought together to place identical songs or songs by the same artist in close proximity. Another way they visualized listening history is by using strings. Strings are simply individual listening sessions ordered by time. (Think of strings as a collection of linked lists.) They then took the strings and combined identical songs into single nodes to form knots. (Knots being stronger because they are actually the same node for a given song, not just duplicate nodes placed nearby.) They used both tangles and knots to create new playlists using route mapping algorithms. By selecting a starting song and ending song and a route efficiency a playlist is generated that can be seen visually as a route between two points. Users can also click to add desired waypoints just as in Google Maps.
Discussion:
I use Last.fm and find it interesting to see my music history displayed graphically. That being said it is a curiosity at best. However, creating playlists by picking a route between two songs from my existing play history is amazing. That could be useful and entertaining. The only changes I would make would be to allow for user definable break. (Sometimes I am listening and say "and now for something completely different." Of course, leaving these in may make the experience more interesting.) A really neat idea and I look forward to seeing it in action.
Summary:
This paper proposes a means of using a visual history to allow a media program to create playlists based on the previous listening sessions of a user. They begin with the prior works, notably Last.fm and Pandora, and move on to the methods they used to allow users to visualize thier listening habits and then create playlists. The first method they use to visual listening history is a tangle. To create a tangle they connect all songs that are listened in succession to each other (directionally). This means that if you listen to "Save A Horse (Ride A Cowboy)" and then "The Rockafeller Skank" and then "The Spirit Of Radio" then they are connected in that order. If you then listen to "The Everglow" and then "The Rockafeller Skank" then the "The Everglow" is connected to a new visual representation of "The Rockafeller Skank." Tangles form in the chain of songs when the different portions are brought together to place identical songs or songs by the same artist in close proximity. Another way they visualized listening history is by using strings. Strings are simply individual listening sessions ordered by time. (Think of strings as a collection of linked lists.) They then took the strings and combined identical songs into single nodes to form knots. (Knots being stronger because they are actually the same node for a given song, not just duplicate nodes placed nearby.) They used both tangles and knots to create new playlists using route mapping algorithms. By selecting a starting song and ending song and a route efficiency a playlist is generated that can be seen visually as a route between two points. Users can also click to add desired waypoints just as in Google Maps.
Discussion:
I use Last.fm and find it interesting to see my music history displayed graphically. That being said it is a curiosity at best. However, creating playlists by picking a route between two songs from my existing play history is amazing. That could be useful and entertaining. The only changes I would make would be to allow for user definable break. (Sometimes I am listening and say "and now for something completely different." Of course, leaving these in may make the experience more interesting.) A really neat idea and I look forward to seeing it in action.
People don't draw straight lines [UIST '08]
Kinematic templates: end-user tools for content-relative cursor manipulations by Richard Fung, University of Waterloo, et al.
Summary:
In this paper the authors introduced Kinematic Templates. They attempt to fill the area between strict objects in images (such as circles) and free hand user drawings. They accomplish this with what they call Kinematic Templates with various behaviors. They defined two types of templates: Passive and Active. Passive merely alter the drawing the user is creating. This changes the pointer movement to drawing ratio to something besides 1:1, depending on how the movement is made and what the template is. Active templates actually move the pointer when the mouse (or pen) remains in the same location. They defined a number of different templates in both categories and allowed the user to set the "strength" of the template. Examples: The sandpaper template (passive) assists in fine details by slowing down drawing in an area. The compass (passive) makes drawings fit to a circle so that with a lesser strength small imperfections in a drawn circle are fixed and in a stronger setting a square mouse movement is drawn almost circularly. The orbit template (active) moves the pointer in a circular motion. They allowed for combining of templates. They combined the orbit with the compass to make spirals. The overall goal was to allow users to have a grey area between rigid shapes and free hand drawings. They conclude they have done fairly well in this in that the user may create more precise and visual appealing drawings that maintain a human element.
Discussion:
This seemed like a really neat tool. It would make a great addition to many drawing programs, particularly those focused on fun and creativity. This paper also seems like a suitable prior work to our second project as it involves computer aided manipulations of a persons hand drawings to fit a preset template. This is something that when you see (and you can on the video) you want to play with (and you can by downloading the GPL'd library and example here).
Summary:
In this paper the authors introduced Kinematic Templates. They attempt to fill the area between strict objects in images (such as circles) and free hand user drawings. They accomplish this with what they call Kinematic Templates with various behaviors. They defined two types of templates: Passive and Active. Passive merely alter the drawing the user is creating. This changes the pointer movement to drawing ratio to something besides 1:1, depending on how the movement is made and what the template is. Active templates actually move the pointer when the mouse (or pen) remains in the same location. They defined a number of different templates in both categories and allowed the user to set the "strength" of the template. Examples: The sandpaper template (passive) assists in fine details by slowing down drawing in an area. The compass (passive) makes drawings fit to a circle so that with a lesser strength small imperfections in a drawn circle are fixed and in a stronger setting a square mouse movement is drawn almost circularly. The orbit template (active) moves the pointer in a circular motion. They allowed for combining of templates. They combined the orbit with the compass to make spirals. The overall goal was to allow users to have a grey area between rigid shapes and free hand drawings. They conclude they have done fairly well in this in that the user may create more precise and visual appealing drawings that maintain a human element.
Discussion:
This seemed like a really neat tool. It would make a great addition to many drawing programs, particularly those focused on fun and creativity. This paper also seems like a suitable prior work to our second project as it involves computer aided manipulations of a persons hand drawings to fit a preset template. This is something that when you see (and you can on the video) you want to play with (and you can by downloading the GPL'd library and example here).
Subscribe to:
Posts (Atom)
