Paper Reading #4: Gestalt

References
Kayur Patel, Naomi Bancroft, Steven M. Drucker, James Fogarty, Andrew J. Ko, and James Landay.  "Gestalt: integrated support for implementation and analysis in machine learning". UIST '10 Proceedings of the 23rd annual ACM symposium on User interface software and technology.  ACM New York, NY, USA ©2010.

 Author Bios
Kayur Patel is a PhD student at the University of Washington specializing in machine learning.

Naomi Bancroft is a Senior undergraduate researcher at the University of Washington. Her interests lie in HCI.

Steven M. Drucker is a  Principal Researcher at Microsoft Research with who specializes in HCI. He received his PhD from MIT.

James Fogarty is currently an assistant professor at UW. His research focuses on HCI and Ubiquitous computing. He received his PhD from Carnegie Melon.

Andrew J.Ko is also currently an assistant professor at UW. His research focuses on the “Human aspects of software development”. He also received his PhD from Carnegie Melon.

James A Landay is a professor at UW, whose research focuses on Automated Usability Evaluation, Demonstrational Interfaces, and Ubiquitous Computing. He received his PhD from Carnegie Melon as well.

Summary 

• Hypothesis - A general purpose Machine Learning tool that allows developers to analyze the information pipeline will lead to greater efficiency and fewer errors.
• Method/Content - The researchers created two problems, one for movie reviews and one for gesture recognition. Eight testers were then given a program for each problem; each program had 5 bugs in it. Within an hour, they were asked to find and fix as many bugs as they could. The tools they used to find and fix these problems were their newly developed Gestalt Framework, and the other was to use a customized version of Matlab. Each participant was asked to solve each problem with each program (4 tests in all).
• Results - The results showed that participants were able to find significantly more errors while using the Gestalt framework. Some even tried to create Gestalt functionality within Matlab. All eight of the users preferred Gestalt over Matlab, and most of them stated that they would likely benefit from using Gestalt in their work. 
• Content - This paper presented Gestalt, which is a new tool for developers of Machine Learning. It then conducted a user study to compare it with other similar software, and found that it was indeed a good tool. It then discussed its strengths and weaknesses. Its main strength lies in its ability for users to view the information pipeline.

 Discussion
Although a general purpose tool cannot necessarily perform all of the same tasks as well as a domain-specific tool, they are often flexible enough to still be a powerful tool. Gestalt seems as if it has a good ways to go before it sees general use, but the results were promising. Throughout the paper, the greatest thing that I saw about the framework was its ability to let you view (and manipulate) the information pipeline. This is key for many applications, especially machine learning. Although their testing methods were not robust, they did serve to show a general sentiment of how Gestalt can be useful to developers.