Gestalt

Authors:
Kayur Patel is a computer science PhD student at University of Washington
Naomi Bancroft is a computer science undergraduate at the University of Washinton
Steven M. Drucker is a Microsoft Researcher
James Fogarty is an assistant professor at the University of Washington
Andrew Ko is also an assistant professor at the University of Washington
James Landay is a professor of computer science at the University of Washington

Summary:
Hypothesis:
The authors believe that they can create a system that applies machine language in a way that is different from traditional programming. It focuses on the learned behavior of a machine rather than describing the behavior of a program like traditional programming.

Methods:
The subjects had to create and run scripts that connected MATLAB and retrieved data from it. They also used the example of analyzing a movie review. They would then decipher a gesture mark. The two main points that were focused on in the paper were sentiment analysis as well as gesture recognition. Because one of the big focuses in this paper was debugging, errors were also hidden within the code and the subjects were suppose to remove them.

Results:
The Gestalt environment provided a much easier way to locate and fix bugs than MATLAB. They found it much easier to operate with the visualization scripting feature.

Contents:
This environment is designed to enhance machine learning. It supports the implementation of a classification pipeline as well as analysis of the data moving through that pipeline. Things such as sentiment and gesture analysis were a huge part in helping this platform be an improvement over the baseline. This is great because these are the two areas in which is has the most practical application in the real world.

Discussion:
These researches attempted, and succeeded at creating a system that would make machine learning debugging less tedious. Because such a large number of subjects preferred the new Gestalt over the baseline I believe they succeeded in their goal. They also have a great new resource in the fields of sentiment analysis and gesture recognition. I think a more expansive paper on this system would be a great thing to read. That way I could get a better idea of a few of the more specific details that went into creating the system and not be limited to simple bug testing.