Paper Reading#4 Gestalt

Title:Gestalt: Integrated Support for

Implementation and Analysis in Machine Learning

Authors: Kayur Patel, Naomi Bancroft, Steven Drucker, James Fogarty, Andrew Ko, James Landay

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

in NY

HCII, Carnegie Mellon University, Microsoft Research WA

Summary:

Hypothesis: Traditional programming allows developers to explicitly describe the behavior of a program, but systems that use machine learning must learn behavior from data.

Although new advanced forms of interactions today they still produced inputs and areas of uncertainty and unpredictability. In such a sense there has been very little progress in terms of such improvements. In order to handle such types problems that are presented in the modern world this paper provides a possible method of basis of dealing with this is a some what convenient fashion. Several demonstrations have been used such as tiny manipulable buttons that can be interacted using touch manipulations or text boxes that can contain several interpretations of various input. Other examples of include methods to handle inputs that have been added in an inaccurate fashion. All these types of things can be used to help provide a better method to reduce uncertainties when interacting with devices today. Explanations contain various levels when it comes to clarifying uncertain aspects of user inputs and explains methods designed in such levels. The basis of such design is mostly designed on probability calculations and they can go through six different types of ambiguous input situations. Through such processes they showed how implementations of user device interaction can be significantly improved.

Contents:

• Gestalt supports the implementation of a classification pipeline, analysis of data as it moves through that pipeline, and easy transitions between implementation and analysis.
• Discussion of Gestalt’s capabilities, including a focus on generalizing lessons from domain-specific tools to
• provide general-purpose support for machine learning.
• An evaluation demonstrating that Gestalt significantly improves developer ability to find and fix bugs in two
• typical applications of machine learning.
• Discussion of current limitations and future opportunities for general-purpose machine learning support.

Discussion

I personally found this whole subject to be quite fascinating and I saw much promise in such aspect. This is especially because I have had many personal problems and experiences that I have had when interacting with devices that this paper speaks of. I think that this paper is a unique tool that can increase productivity greatly.