LeGo 2008: From Local Patterns to Global Models

Participants Survey

[.pdf |.doc |.xls response form]

 

 

 

This workshop deals with techniques that use local pattern mining for improving global modeling. We subsume such techniques under the LeGo process model with three interacting phases: local pattern mining, pattern set selection, and global modeling. Please refer to the first paper in the proceedings for more details and some of the basic terminology used.

As part of our workshop program, we have put together a participants survey, the results of which we intend to present at the end of the day. We would like to ask you to take a look at this questionnaire, and tell us about your views. The questions concern the topic of the workshop.

If possible, please use this .xls response file to express your views (although we will also accept all formats). In this file, simply change any 0 into a 1 for the options that most fit your opinion. If you have no specific preference for some questions, you can leave all options 0. However, we encourage you to answer all questions, even if there is no clear-cut answer. For some options, you are asked to provide free-text. Additionally, you can add as many comments behind every option as you wish. The first two questions are optional. If you leave these blank, we will do our best to preserve anonymity, and will not store your email address.

You can email your response to LeGo-08@ke.informatik.tu-darmstadt.de. We need to receive your feedback before Wednesday September 10, at 10:00 am Central European Time, in order to process it.

Please feel free to forward the survey to colleagues that might be interested in the subject, and ask them to send in their opinions as well. Participation in the survey is not restricted to Workshop attendees!


1.      Name of participant (optional)

2.      Affiliation (optional)

3.      The Global Modeling phase can be straightforward. All the complexity and representational power is located in the Pattern Discovery phase.

A.                I agree.

B.                I disagree.

 

4.      Pattern Subset Selection will make the final global model…

A.                more accurate, because the global model is not hampered by large numbers of redundant patterns/features.

B.                less accurate, because pattern selection results in loss of information.

C.                about as accurate as a model induced from all patterns.

 

5.      Pattern Subset Selection is only useful if a small set of important patterns is needed for human inspection.

A.                I agree.

B.                I disagree.

 

6.      Compared to a traditionally induced model, a LeGo-based global model will be… (multiple answers possible)

A.                better, because of the exploratory (and often complete) search in the Pattern Discovery phase.

B.                worse because of the loss of information in binary patterns/features.

C.                worse because of the risk of overfitting and multiple testing implicit in the Pattern Discovery phase.

D.                better, but only in complex and structured domains, such as multi-relational and graphical.

 

7.      The workings and parameters of the three LeGo phases can best be

A.                treated as independent. Greater flexibility allows for different algorithms to be plugged in.

B.                tuned to each other. By adapting the settings to the previous and next phase, optimal results can be achieved.

C.                It depends on the application.

 

8.      I view LeGo as… (multiple answers possible)

A.                … the ideal solution for building good models

B.                a good approach for doing something useful with the patterns I have found.

C.                a conceptual framework for thinking and talking about different mining and modeling steps.

D.                an idealized and overly simplified model that doesn’t really fit the complex applications I work with.

E.                 a new name for an old trick.

F.                 Other. (please specify)

 

9.      The so-called covering-approach (a.k.a. Separate & Conquer) is not an instance of Lego, as the patterns are not all discovered a priori.

A.                I agree.

B.                I disagree.

 

10. I will be working on… (multiple answers possible)

A.                A framework for global modeling from local patterns.

B.                Local pattern discovery.

C.                Pattern subset selection.

D.                Metrics for pattern subset selection.

E.                 Efficient implementations of pattern subset selection.

F.                 Global modeling.

G.                Global modeling from sets of patterns.

H.                Propagation of constraints: translating global constraints on the model into constraints of the local patterns.

I.                   General purpose constraints: finding local patterns that are useful for a variety of global models.

J.                   Parallel Universes.

K.                Not-LeGo related subjects.

L.                 Other. (please specify)

 

11. We should be working on… (multiple answers possible)

A.                A framework for global modeling from local patterns.

B.                Local pattern discovery.

C.                Pattern subset selection.

D.                Metrics for pattern subset selection.

E.                 Efficient implementations of pattern subset selection.

F.                 Global modeling.

G.                Global modeling from sets of patterns.

H.                Propagation of constraints: translating global constraints on the model into constraints of the local patterns.

I.                   General purpose constraints: finding local patterns that are useful for a variety of global models.

J.                   Parallel Universes.

K.                Not-LeGo related subjects.

L.                 Other. (please specify)

 

12. There should again be a LeGo workshop next year.

A.                I agree. There is still a lot of research to be done in this area.

B.                I disagree. Most of the problems have been solved.

C.                I don’t care either way.

 

13. I think it is a good time for a publication on this subject (not necessarily limited to the workshop results).

A.                Yes, an edited volume.

B.                 Yes, a journal special issue.

C.                 No.

 

14. I would be interested in contributing to a publication on this subject.

A.                An edited volume.

B.                 A journal special issue.

C.                 Either.

D.                Neither.

 

15. General comments and suggestions: