Knowledge discovery at Purdue
In a combination of KM and chemical engineering, I came across the abstract below in my ACM News Service email from last week. I had seen the title, but kept delaying hunting down the link. It turns out that a few of my friends at Purdue are involved in this project.
In case you don't want to follow it yourself, their research combines high-tech visualization techniques with a number of analysis techniques to enable researchers to test theories on massive, complex data sets. Data mining on steroids, something my former advisor has turned to as well. Yum.
Purdue University researchers have created a new computer-aided product design method that utilizes supercomputers, artificial intelligence, and large 3D displays. Purdue chemical engineering professor W. Nicholas Delgass explains that conventional computer-aided discovery is focused on data-mining where a small piece of relevant data is targeted; this model is well-suited for some tasks, but a better method is called "knowledge discovery," a process Delgass likens to sifting through a warehouse of mechanical parts and piecing together a complex instrument. In recent years, high-throughput experimentation--where researchers conduct thousands of small-scale experiments in a short time-span--has led to a data glut requiring this new knowledge discovery approach. The Purdue system immerses scientists in data and allows them to interact using the vernacular of their particular field. Neural networks and pre-loaded rules-of-thumb help create "forward models" scientists can use to find a particular type of molecule that provides needed characteristics, for example; genetic algorithms also play a role, helping to refine these models based on Darwinian selection. The system also makes use of supercomputing capability to quickly process complex models such as chemical reactions and visualize the results on a tiled wall, or a 12-foot-wide, seven-foot-high bank of integrated displays. Users wear special glasses that provide images in stereo, with slightly different images projected for the left and right eyes. The immersive environment and simulated experiments help scientists quickly determine which course to pursue and make the scientific process much more efficient. The multidisciplinary effort to develop the method included researchers from Purdue's School of Technology, Information Technology at Purdue, School of Science, College of Engineering, and the e-Enterprise Center in Purdue's Discovery Park, and originally began in 1988 with funding from the National Science Foundation.
Previous entry: Embedded knowledge
Next entry: Best practices from APQC