Blog #4 Gestalt

Authors: 

-Kayur Patel - PhD Student at Washington University

-Naomi Bancroft - currently working for Google

-Steven M. Drucker - PR at Microsoft Research

-Andrew J. Ko - Assistant professor at the University of Washington

-James A. Landay - Profesor of CSCE at the University of Washington

-James Fogarty - Assistant professor at the University of Washington

Presentation Venue: 

UIST’10, October 3–6, 2010, New York, New York, USA.

Summary:

The authors hypothesized that programmers like to design programs to behave in a certain way, but that current machine learning systems must be taught behaviors.

The authors tested their hypothesis by creating a developing environment for their test subjects develop in. The testers were given API's that they could use all of Gestalt's visualizations. The baseline test and Gestalt used the same data structures to hold all of the data that each of the users wanted to use. However the baseline test the users had to write their own code to connect data, attributes, and classifications. The users were told that the solutions given to them had 5 bugs, and they must find them. The test subjects all prefered Gestalt to the baseline. They found it much easier to locate and fix the bugs using Gestalt.

Discussion:

The authors research was very well done and was methodically planned. I liked how quickly the test subjects were able to find bugs in when using Gestalt vs the basline. I know how frustrating it is to work with machine learning algorithms. It is very difficult to troubleshoot them when something is going wrong. I would have like to try out Gestalt and see exactly how it works.