Blog Paper Reading #4: Gestalt

References

Authors:
Kayur Patel     University of Washington, Seattle, WA, USA
Naomi Bancroft     University of Washington, Seattle, WA, USA
Steven M. Drucker     Microsoft Research, Seattle, WA, USA
James Fogarty     University of Washington, Seattle, WA, USA
Andrew J. Ko     University of Washington, Seattle, WA, USA
James Landay     University of Washington, Seattle, WA, USA

Published in:
UIST '10 Proceedings of the 23nd annual ACM symposium on User interface software and technology

Summary

This paper is about the development environment: "Gestalt."  Gestalt is a program designed to aid in 'machine learning.'  Machine learning is the process of designing algorithms that focus on capturing behavioural patterns through the processing of data. The authors recognize the difference between the two tasks of implementing and analyzing data pipelines in their paper and explain how Gestalt differentiates between the two and allows users to switch back and forth between these tasks easily.

The authors explain how many different kinds of problems can be solved using similar general methods with machine learning and Gestalt.  Their primary examples are determining whether a movie review is negative or positive by analyzing the vocabulary and grammar of the review and recognizing hand and pen gestures.

For the authors, they conclude that the best method to help develops utilize machine learning is by exposing the entire data pipeline.  They reference other machine learning tools that either attempt to simplify or expedite machine learning by hiding some steps in the pipeline from the users, but the authors conclude that this only hinders the developer and thus, they choose to make Gestalt show the entire process.

The rest of the paper goes into detail about Gestalt itself and how it present its data to the users. 


Discussion

 I thought this paper was actually pretty interesting.  At first, the concept of using pure data to drive programming seemed foreign and odd to me, but after reading about the uses I began to see the benefits of such a system.  Their overall goal and research method seemed sound to me and I really didn't question any of it.  Machine learning seems like such a great idea, I just think it needs to be applied only in certain areas.