Paper Title: Gestalt: Integrated Support for Implementation and Analysis in Machine Learning
Authors: Kayur Patel, Naomi Bancroft, Steven Drucker, James Fogarty, Andrew Ko & James Landay
Author Bios:
Kayur Patel is a student at the University of Washington pursuing a Ph.D. in Computer Science. He is a part of the DuB group and is being advised by James Fogarty and James Landay. He was previously supported by the US Government in the form of a Natioal Defense Science and Engineering Graduate (NDSEG) Fellowship.
Naomi Bancroft is an undergraduate student in the Computer Science & Engineering department at the University of Washington. She is currently doing research with Kayur Patel and James Fogarty on Human-Computer Interaction. She is a pursuing a double major in Computer Science and Linguistics.
Steven Drucker is a Principal Scientist at the LiveLabs Group at Microsoft. He is working on user interaction and information visualization for web based projects. He is also an affiliate professor at the University of Washington. Previously, he was a researcher at Microsoft Research for 11 years. First, as a lead researcher for the Next Media Research Group in Microsoft for 6 years where he examined how the addition of user interaction transforms conventional media.
James Fogarty is an Assistant Professor in the Department of Computer Science & Engineering at the University of Washington. He is broadly interested in Human-Computer Interaction, User Interface Software & Technology and Ubiquitous Computing. My research focuses on developing, deploying, and evaluating new approaches to the human obstacles surrounding the widespread adoption of ubiquitous and intelligent computing technologies.
Andrew Ko is an Assistant Professor at the Information School at the University of Washington. His research interests include social and cognitive factors in software engineering, end user software engineering, user interface software and technology, and programming language design.
James Landay is an Associate Professor in Computer Science & Engineering at the University of Washington, specializing in human-computer interaction. From 2003 through 2006 he was also the Laboratory Director of Intel Research Seattle, a university affiliated research lab exploring ubiquitous computing.
Presentation Venue: The presentation of this paper took place at UIST '10 proceedings of the 23rd annual ACM symposium on User Interface Software and Technology in New York.
Summary Section:
Paper Hypothesis: The hypothesis of the paper is to test the Gestalt tool. The authors test if Gestalt allows developers to implement a classification pipeline, analyze data as it moves through that pipeline, and can easily transition between implementation and analysis. They have focused on 5 major points in this paper:
1) Providing Structure While Maintaining Flexibility
2) Appropriate Data Structures
3) Visualizing and Aggregating Examples
4) Interactive, Connected Visualizations
5) The "Gestalt" of Gestalt
How the hypothesis was tested: Gestalt is compatible with most of the modern development environments like Eclipse, Microsoft Visual Studio etc). They tested the hypothesis by evaluating bug finding in Gestalt. They compared the bug-finding performance for participants using Gestalt with a baseline condition similar to MATLAB.
Result of the hypothesis: The participants unanimously preferred Gestalt and were able to find and fix more bugs using Gestalt than using the baseline.
General Summary: This paper presents Gestalt, a general-purpose tool for applying machine learning. Gestalt targets developers, providing full support for writing code to specify the series of steps in a classification pipeline. In supporting a wide range of classification problems, Gestalt generalizes the lessons of prior domain-specific tools.
Discussion:
Significance of the paper: This paper provides a great motivation for users to create their own individual example visualizations. Also, Gestalt's connected visualizations make it easy to compare the data, attributes, and classification results of users.
Faults of the work: Gestalt has several limitations like the point where its tasks has pipelines that could be run in real-time. This can be proven expensive for many learning problems. Also Gestalt is not very effective in finding bugs in unfamiliar code.
Interesting future work: Gestalt is an effective and a relatively efficient tool that can help developers understand relationships between the various steps in a classification pipeline and can generalize advances from prior domain-specific tools to provide general-purpose support.
Authors: Kayur Patel, Naomi Bancroft, Steven Drucker, James Fogarty, Andrew Ko & James Landay
Author Bios:
Kayur Patel is a student at the University of Washington pursuing a Ph.D. in Computer Science. He is a part of the DuB group and is being advised by James Fogarty and James Landay. He was previously supported by the US Government in the form of a Natioal Defense Science and Engineering Graduate (NDSEG) Fellowship.
Naomi Bancroft is an undergraduate student in the Computer Science & Engineering department at the University of Washington. She is currently doing research with Kayur Patel and James Fogarty on Human-Computer Interaction. She is a pursuing a double major in Computer Science and Linguistics.
Steven Drucker is a Principal Scientist at the LiveLabs Group at Microsoft. He is working on user interaction and information visualization for web based projects. He is also an affiliate professor at the University of Washington. Previously, he was a researcher at Microsoft Research for 11 years. First, as a lead researcher for the Next Media Research Group in Microsoft for 6 years where he examined how the addition of user interaction transforms conventional media.
James Fogarty is an Assistant Professor in the Department of Computer Science & Engineering at the University of Washington. He is broadly interested in Human-Computer Interaction, User Interface Software & Technology and Ubiquitous Computing. My research focuses on developing, deploying, and evaluating new approaches to the human obstacles surrounding the widespread adoption of ubiquitous and intelligent computing technologies.
Andrew Ko is an Assistant Professor at the Information School at the University of Washington. His research interests include social and cognitive factors in software engineering, end user software engineering, user interface software and technology, and programming language design.
James Landay is an Associate Professor in Computer Science & Engineering at the University of Washington, specializing in human-computer interaction. From 2003 through 2006 he was also the Laboratory Director of Intel Research Seattle, a university affiliated research lab exploring ubiquitous computing.
Presentation Venue: The presentation of this paper took place at UIST '10 proceedings of the 23rd annual ACM symposium on User Interface Software and Technology in New York.
Paper Hypothesis: The hypothesis of the paper is to test the Gestalt tool. The authors test if Gestalt allows developers to implement a classification pipeline, analyze data as it moves through that pipeline, and can easily transition between implementation and analysis. They have focused on 5 major points in this paper:
1) Providing Structure While Maintaining Flexibility
2) Appropriate Data Structures
3) Visualizing and Aggregating Examples
4) Interactive, Connected Visualizations
5) The "Gestalt" of Gestalt
How the hypothesis was tested: Gestalt is compatible with most of the modern development environments like Eclipse, Microsoft Visual Studio etc). They tested the hypothesis by evaluating bug finding in Gestalt. They compared the bug-finding performance for participants using Gestalt with a baseline condition similar to MATLAB.
Result of the hypothesis: The participants unanimously preferred Gestalt and were able to find and fix more bugs using Gestalt than using the baseline.
General Summary: This paper presents Gestalt, a general-purpose tool for applying machine learning. Gestalt targets developers, providing full support for writing code to specify the series of steps in a classification pipeline. In supporting a wide range of classification problems, Gestalt generalizes the lessons of prior domain-specific tools.
Discussion:
Significance of the paper: This paper provides a great motivation for users to create their own individual example visualizations. Also, Gestalt's connected visualizations make it easy to compare the data, attributes, and classification results of users.
Faults of the work: Gestalt has several limitations like the point where its tasks has pipelines that could be run in real-time. This can be proven expensive for many learning problems. Also Gestalt is not very effective in finding bugs in unfamiliar code.
Interesting future work: Gestalt is an effective and a relatively efficient tool that can help developers understand relationships between the various steps in a classification pipeline and can generalize advances from prior domain-specific tools to provide general-purpose support.
No comments:
Post a Comment