Paper Reading #4 - Gestalt

• Paper Title: Gestalt: Integrated Support for Implementation and Analysis in Machine Learning
• Authors:
• Kayur Patel, Naomi Bancrof, James Fogarty, and James A. Landa of Computer Science and Engineering DUB Group, University of Washington
•  Steven M. Drucker of Microsoft Research
•  Andrew J. K of Information School DUB Group, University of Washington
• Presented at 23rd annual ACM symposium on User interface software and technology, New York, NY, 2010

See paper here: http://dl.acm.org/citation.cfm?doid=1866029.1866038
For information on machine learning: http://en.wikipedia.org/wiki/Machine_learning

Summary

     This paper discusses Gestalt, a development environment for applications of machine learning (an AI process where a machine makes some kind of decision based on given data). Gestalt is designed to simplify the process of creating machine learning applications by giving the developer easy access to both the implementation and analysis sides of the development process. Patel et al. give two specific problems dealing with machine learning; sentiment analysis (a way of categorizing text) and gesture recognition (as with pen-touch input). To test the effectiveness of Gestalt, the researchers designed two applications, each implementing one of two given problems, and intentionally left bugs in them. They then had participants debug the applications using Gestalt and another "baseline" program and describe which method was preferred. The study showed a unanimous preference toward Gestalt, with participants particularly praising the ease of transferring between implementation and analysis.

My View

     Since machine learning is not a field I am particularly familiar with, this paper was a bit more difficult for me to understand than some of the others I've written about. From what I have gathered, however, it appears as though this application is the Visual Studio, if you will, of machine learning applications. For those unfamiliar with Visual Studio, it is a programming interface that will assist you in building your program and offers powerful debugging tools  that will allow you to go through each and every step and view the changes.
     If I am correct in this comparison, I could see Gestalt being a very useful tool in building and debugging programs that involve machine learning. As I said, I am not experienced in creating machine learning applications, but I have felt the frustration of not being able to access all the information I feel I need to practically debug a program, and any interface that allows me to actively see more information gets a thumbs up from me.