My research interests have meandered quite a bit over the last several years. Below is a list of the projects on which I have worked, including short descriptions for several. If you would like to know more about them, please send me email: corin@cs.washington.edu. Or, take a look at my publications page for published materials.
Most web sites today are designed with a "one-size-fits-all" philosophy. The web site designer "specs out" the needs of the web visitor, and then creates the content accordingly. However, one size frequently does not fit all. In particular, an increasingly many web visitors are browsing from mobile devices -- wireless PDAs and web-enabled cell phones. These devices have extremely small screens and slow network connections, and content optimized for desktop PCs is nearly impossible to use.
To best serve the needs of this growing community, we propose building web site personalizers that observe the behavior of web visitors and automatically customize and adapt sites for each individual mobile visitor. Our approach to personalization employs a utility-maximizing search through the space of personalized web sites. Following this framework we have implemented a personalizer: Proteus. Proteus allows changes to site navigation (adding or removing links) as well as content manipulation (rearranging or eliding content), and evaluates the result with a learned model of the current visitor. Preliminary experiments indicate that web site personalizers can significantly improve the web experience for mobile visitors.
See also my Adaptive Web Sites site.
Early research in web site management identified a key principle: the separation of data management, site structure, and page presentation. This separation was made through the introduction of a logical representation, the site graph, defined as a view over underlying data. While the separation of these three tasks provides many benefits, existing systems require that the user implement the web site with the same system they use to design it, and this restriction is problematic for three reasons. First, many users are familiar with and prefer another implementation tool. Second, other tools may provide useful, new, or specialized features. Third, a complex site may be controlled by multiple organizations whose standards require different tools. In this work we present a new architecture for declarative web-site management that separates design and implementation. Following this architecture, we implemented Tiramisu, and used it to design and build several web site.
Sensory Graphplan (SGP) is a sound, complete planner based on Blum and Furst's Graphplan. SGP includes support for conditional effects, universal and existential quantification, uncertainty, sensing actions, and uses the PDDL language syntax. SGP is written in Lisp, is efficient, and is well-suited for use in the classroom or research.
Computers have traditionally been used to produce photorealistic images -- pictures that look just like you were there. A wholly different approach to producing imagery is non-photorealistic rendering, in which the goal is to produce an image that's representative, but not an exact picture, of a real object. Pen-and-ink illustration is one example of non-photorealistic rendering.
I began work with computer-generated pen-and-ink illustration after others at the UW had broken the ground. The specific work I with which I was involved concerned storing and transmitting pen-and-ink illustrations. In a nutshell, the problem was that pen-and-ink illustrations were pretty at high resolution (for instance, 5" x 7 " at 300dpi), but such hi-res graphic files (or PostScript files) were prohibitively expensive to store or send on the network. Instead, our approach was to represent pen-and-ink illustrations as an underlying image file (which could easily be compressed both in resolution and file size) and the inked textures that would be used.
The color of an object that we see depends not only on the object's material and the available light, but also on the surrounding objects' colors. For instance, a grey square on a black background will actually look brighter than the same grey square on a white background. This effect is know as chromatic induction -- the colors of surrounding objects are inducing an effect on the color of the central object.
In this research, we sought to better understand chromatic induction, and to in fact take advantage of the effect to produce more visually pleasing images. The motivating task was to automatically adjust the font color on a slide of text whose background was a gradient from light to dark, so that the font would be perceived as a uniform color. Unfortunately, we didn't get this far (or, in fact, very close).