LCS & GBML Central – Page 6557 – The LCS and GBML community stop

CIG-2009 proceedings available online

Pier Luca Lanzi just sent a note saying saying that the proceedings of the 2009 Symposium on Computational Intelligence and Games (CIG-2009) are now available on-line at
http://www.ieee-cig.org/cig-2009/Proceedings/
Related Posts

Pier Luca Lanzi just sent a note saying saying that the proceedings of the 2009 Symposium on Computational Intelligence and Games (CIG-2009) are now available on-line at

http://www.ieee-cig.org/cig-2009/Proceedings/

Liquid: RDF endpoint for FluidDB

A while ago I wrote some thoughts about how to map RDF to and from FluidDB. There I explored how you could map RDF onto FluidDB, and how to get it back. That got me thinking about how to get a simple endpoint you could query for RDF. Imagine that you could pull FluidDB data in RDF, then I could just get all the flexibility of SPARQL for free. With this idea in my mind I just went and grabbed Meandre, the JFLuidDB library started by Ross Jones, and build a few components.

The main goal was to be able to get an object, list of the tags, and express the result in RDF. FluidDB helps the mapping since objects are uniquely identified by URIs. For instance, the unique object 5ff74371-455b-4299-83f9-ba13ae898ad1 (FluidDB relies on UUID version four with the form xxxxxxxx-xxxx-4xxx-yxxx-xxxxxxxxxxxx) is uniquely identified by http://sandbox.fluidinfo.com/objects/5ff74371-455b-4299-83f9-ba13ae898ad1 (or a url of the form http://sandbox.fluidinfo.com/objects/xxxxxxxx-xxxx-4xxx-yxxx-xxxxxxxxxxxx), in case you are using the sandbox or http://fluiddb.fluidinfo.com/objects/5ff74371-455b-4299-83f9-ba13ae898ad1 if you are using the main instance. Same story for tags. The tag fluiddb/about can be uniquely identified by the URI http://sandbox.fluidinfo.com/tags/fluiddb/about, or http://fluiddb.fluidinfo.com/tags/fluiddb/about.

A simple RDF description for and object

Once you get the object back the basic translated RDF version for object a10ab0f3-ef56-4fc0-a8fa-4d452d8ab1db should look like as the listing below in TURTLE notation.

<http://sandbox.fluidinfo.com/objects/a10ab0f3-ef56-4fc0-a8fa-4d452d8ab1db>
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#type>
              <http://sandbox.fluidinfo.com/objects/> , <http://www.w3.org/1999/02/22-rdf-syntax-ns#Bag> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_1>
              <http://sandbox.fluidinfo.com/tags/fluiddb/about> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_2>
              <http://sandbox.fluidinfo.com/tags/fluiddb/tags/path> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_3>
              <http://sandbox.fluidinfo.com/tags/fluiddb/tags/description> ;
      <http://purl.org/dc/elements/1.1/description>
              "Object for the attribute fluiddb/default/tags/permission/update/policy"^^<http://www.w3.org/2001/XMLSchema#string> .

I will break the above example into small chunks and explain the above example into the three main pieces involved (the id, the about, and the tags). The basic construct is simple. First a triple to mark the object as a FluidDB object.

<http://sandbox.fluidinfo.com/objects/a10ab0f3-ef56-4fc0-a8fa-4d452d8ab1db>
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#type>
              <http://sandbox.fluidinfo.com/objects/>   
.

Then if the object has an about associated on creation, another triple gets generated and added, as shown below. To be consistent, I suggest reusing DC description since that is what the about for an object tend to indicate.

<http://sandbox.fluidinfo.com/objects/a10ab0f3-ef56-4fc0-a8fa-4d452d8ab1db>
      <http://purl.org/dc/elements/1.1/description>
              "Object for the attribute fluiddb/default/tags/permission/update/policy"^^<http://www.w3.org/2001/XMLSchema#string> 
.

Finally, if there are tags associated to the object, a bag gets created, and all the URI describing the tags get pushed into the bag as shown below.

<http://sandbox.fluidinfo.com/objects/a10ab0f3-ef56-4fc0-a8fa-4d452d8ab1db>
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#type>
              <http://www.w3.org/1999/02/22-rdf-syntax-ns#Bag> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_1>
              <http://sandbox.fluidinfo.com/tags/fluiddb/about> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_2>
              <http://sandbox.fluidinfo.com/tags/fluiddb/tags/path> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_3>
              <http://sandbox.fluidinfo.com/tags/fluiddb/tags/description>
.

Creating and RDF endpoint

Armed with the previous, the thing should be easy. Just allow querying for objects, then collect the object information, and finally generate the final RDF. Using Meandre and JFLuidDB I wrote a few components that allow the simple creation of such an endpoint as illustrated by the picture below.

The basic mechanism is simple. Just push the query into the Query for objects component. This component will stream each of the uuid of the matched objects to Read object which pulls the object information. Then the object is passed to Object to RDF model that basically generates the RDF snipped shown in the example shown above for each of the objects pushed. Finally all the RDF fragments are reduced together by component Wrapped models reducer. Then the resulting RDF model just gets serialize into text using the Turtle notation. Finally the serialized text is printed to the console. The equivalent code could be express as a ZigZag script as:

#
# Imports eliminated for clarity
#

#
# Create the component aliases
#
alias  as OBJECT_TO_RDF
alias  as PRINT_OBJECT
alias  as QUERY_FOR_OBJECTS
alias  as READS_THE_REQUESTED_OBJECT
alias  as WRAPPED_MODELS_REDUCER
alias  as MODEL_TO_RDF_TEXT
alias  as PUSH_STRING

#
# Create the component instances
#
push_query_string = PUSH_STRING()
wrapped_models_reducer = WRAPPED_MODELS_REDUCER()
query_for_objects = QUERY_FOR_OBJECTS()
reads_object = READS_THE_REQUESTED_OBJECT()
model_to_rdf_text = MODEL_TO_RDF_TEXT()
print_rdf_text = PRINT_OBJECT()
object_to_rdf_model = OBJECT_TO_RDF()

#
# Set component properties
#
push_query_string.message = "has fluiddb/tag/path"
query_for_objects.fluiddb_url = "http://sandbox.fluidinfo.com"
eads_object.fluiddb_url = "http://sandbox.fluidinfo.com"
model_to_rdf_text.rdf_dialect = "TTL"

#
# Create the flow by connecting the components
#
@query_for_objects_outputs = query_for_objects()
@model_to_rdf_text_outputs = model_to_rdf_text()
@push_query_string_outputs = push_query_string()
@object_to_rdf_model_outputs = object_to_rdf_model()
@reads_object_outputs = reads_object()
@wrapped_models_reducer_outputs = wrapped_models_reducer()

query_for_objects(text: push_query_string_outputs.text)
model_to_rdf_text(model: wrapped_models_reducer_outputs.model)
object_to_rdf_model(object: reads_object_outputs.object)
reads_object(uuid: query_for_objects_outputs.uuid)[+200!]
print_rdf_text(object: model_to_rdf_text_outputs.text)
wrapped_models_reducer(model: object_to_rdf_model_outputs.model)

The only interesting element in the script is the [+200!] entry that creates 200 parallel copies of read object that will concurrently hit FluidDB to pull the data, trying to minimize the latency. The script could be compiled into a MAU and run. The output of the execution would look like the following:

$ java -jar zzre-1.4.7.jar pull-test.mau 
Meandre MAU Executor [1.0.1vcli/1.4.7]
All rights reserved by DITA, NCSA, UofI (2007-2009)
THIS SOFTWARE IS PROVIDED UNDER University of Illinois/NCSA OPEN SOURCE LICENSE.
 
Executing MAU file pull-test.mau
Creating temp dir pull-test.mau.run
Creating temp dir pull-test.mau.public_resources
 
Preparing flow: meandre://seasr.org/zigzag/1253813636945/4416962494019783033/flow/pull-test-mau/
2009-09-24 12:34:38.480::INFO:  jetty-6.1.x
2009-09-24 12:34:38.495::INFO:  Started SocketConnector@0.0.0.0:1715
Preparation completed correctly
 
Execution started at: 2009-09-24T12:34:38
----------------------------------------------------------------------------
<http://sandbox.fluidinfo.com/objects/a24b4a18-5483-47c6-9b62-0955210c7ebd>
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#type>
              <http://sandbox.fluidinfo.com/objects/> , <http://www.w3.org/1999/02/22-rdf-syntax-ns#Bag> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_1>
              <http://sandbox.fluidinfo.com/tags/fluiddb/about> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_2>
              <http://sandbox.fluidinfo.com/tags/fluiddb/tags/path> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_3>
              <http://sandbox.fluidinfo.com/tags/fluiddb/tags/description> ;
      <http://purl.org/dc/elements/1.1/description>
              "Object for the attribute test/Net::FluidDB-name-1253772095.82845-0.944567286499904"^^<http://www.w3.org/2001/XMLSchema#string> .
 
<http://sandbox.fluidinfo.com/objects/5ff74371-455b-4299-83f9-ba13ae898ad1>
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#type>
              <http://sandbox.fluidinfo.com/objects/> , <http://www.w3.org/1999/02/22-rdf-syntax-ns#Bag> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_1>
              <http://sandbox.fluidinfo.com/tags/fluiddb/about> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_2>
              <http://sandbox.fluidinfo.com/tags/fluiddb/tags/path> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_3>
              <http://sandbox.fluidinfo.com/tags/fluiddb/tags/description> ;
      <http://purl.org/dc/elements/1.1/description>
              "Object for the attribute test/Net::FluidDB-name-1253622685.3231461-0.437099602163897316"^^<http://www.w3.org/2001/XMLSchema#string> .
 
<http://sandbox.fluidinfo.com/objects/67e52346-527e-4bb7-b8f3-05fa8a8ae35b>
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#type>
              <http://sandbox.fluidinfo.com/objects/> , <http://www.w3.org/1999/02/22-rdf-syntax-ns#Bag> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_1>
              <http://sandbox.fluidinfo.com/tags/fluiddb/about> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_2>
              <http://sandbox.fluidinfo.com/tags/fluiddb/tags/path> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_3>
              <http://sandbox.fluidinfo.com/tags/fluiddb/tags/description> ;
      <http://purl.org/dc/elements/1.1/description>
              "Object for the attribute test/Net::FluidDB-name-1253620190.69175-0.861614257420541"^^<http://www.w3.org/2001/XMLSchema#string> .
 
<http://sandbox.fluidinfo.com/objects/8a65a184-03d9-4881-95df-02fa0561a86f>
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#type>
              <http://sandbox.fluidinfo.com/objects/> , <http://www.w3.org/1999/02/22-rdf-syntax-ns#Bag> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_1>
              <http://sandbox.fluidinfo.com/tags/fluiddb/about> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_2>
              <http://sandbox.fluidinfo.com/tags/fluiddb/tags/path> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_3>
              <http://sandbox.fluidinfo.com/tags/fluiddb/tags/description> ;
      <http://purl.org/dc/elements/1.1/description>
              "Object for the attribute fluiddb/namespaces/permission/update/exceptions"^^<http://www.w3.org/2001/XMLSchema#string> .
 
<http://sandbox.fluidinfo.com/objects/335b44e9-a72f-479d-ad60-3661a35231ba>
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#type>
              <http://sandbox.fluidinfo.com/objects/> , <http://www.w3.org/1999/02/22-rdf-syntax-ns#Bag> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_1>
              <http://sandbox.fluidinfo.com/tags/fluiddb/about> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_2>
              <http://sandbox.fluidinfo.com/tags/fluiddb/tags/path> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_3>
              <http://sandbox.fluidinfo.com/tags/fluiddb/tags/description> ;
      <http://purl.org/dc/elements/1.1/description>
              "Object for the attribute test/Net::FluidDB-name-1253776141.95577-0.284175700598524"^^<http://www.w3.org/2001/XMLSchema#string> .
 
<http://sandbox.fluidinfo.com/objects/3bbf1cc6-731c-4e56-a664-adeb5484334f>
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#type>
              <http://sandbox.fluidinfo.com/objects/> , <http://www.w3.org/1999/02/22-rdf-syntax-ns#Bag> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_1>
              <http://sandbox.fluidinfo.com/tags/fluiddb/about> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_2>
              <http://sandbox.fluidinfo.com/tags/fluiddb/tags/path> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_3>
              <http://sandbox.fluidinfo.com/tags/fluiddb/tags/description> ;
      <http://purl.org/dc/elements/1.1/description>
              "Object for the attribute fluiddb/namespaces/permission/delete/policy"^^<http://www.w3.org/2001/XMLSchema#string> .
 
<http://sandbox.fluidinfo.com/objects/aba5adcf-fd44-40ab-b702-9cc635650bc3>
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#type>
              <http://sandbox.fluidinfo.com/objects/> , <http://www.w3.org/1999/02/22-rdf-syntax-ns#Bag> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_1>
              <http://sandbox.fluidinfo.com/tags/fluiddb/about> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_2>
              <http://sandbox.fluidinfo.com/tags/fluiddb/tags/path> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_3>
              <http://sandbox.fluidinfo.com/tags/fluiddb/tags/description> ;
      <http://purl.org/dc/elements/1.1/description>
              "Object for the attribute test/Net::FluidDB-name-1253614713.757-0.604769721717702"^^<http://www.w3.org/2001/XMLSchema#string> .
 
<http://sandbox.fluidinfo.com/objects/f61ceb3b-33df-4356-8e7d-c56d3d0ae338>
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#type>
              <http://sandbox.fluidinfo.com/objects/> , <http://www.w3.org/1999/02/22-rdf-syntax-ns#Bag> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_1>
              <http://sandbox.fluidinfo.com/tags/fluiddb/about> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_2>
              <http://sandbox.fluidinfo.com/tags/fluiddb/tags/path> ;
      <http://www.w3.org/1999/02/22-rdf-syntax-ns#_3>
              <http://sandbox.fluidinfo.com/tags/fluiddb/tags/description> ;
      <http://purl.org/dc/elements/1.1/description>
              "Object for the attribute test/Net::FluidDB-name-1253615887.80879-0.0437609496034099"^^<http://www.w3.org/2001/XMLSchema#string> .
 
...

That’s it! A first RDF dump of the query!

The not so great news

The current FluidDB API does not provide any method to be able to pull data from more than one object at once. That basically means, that for each uuid a call to the server needs to be process. That is a huge latency overhead. The FluidDB guys know about it and they are scratching their heads on how to provide a “multi get”. A full trace of the output can be found on this FluidDB RDF endpoint trace.

This element is crucial for any RDF endpoint. Above I left out a basic element, the time measures. That part looks like:

Flow execution statistics

Flow unique execution ID : meandre://seasr.org/zigzag/1253813636945/4416962494019783033/flow/pull-test-mau/8D8E354A/1253813678323/1493255769/
Flow state               : ended
Started at               : Thu Sep 24 12:34:38 CDT 2009
Last update              : Thu Sep 24 12:37:28 CDT 2009
Total run time (ms)      : 170144

Basically 170s to pull only 238 objects, where all the time is spent round tripping to FluidDB.

Getting there

This basically means that such high latency would not allow efficient interactive usage of the end point. However, this exercise was useful to prof that simple RDF endpoints for FluidDB are possible and would greatly boost the flexibility of interaction with FluidDB . The current form of the endpoint is may still have value if you are not in a hurry, allowing you to run SPARQL queries against FluidDB data and get the best of both worlds.

The code use

If you are interested on running the code, you may need Meandre and the components I put together for the experiment, that you can get from http://github.com/xllora/liquid.

GP challenge: evolving energy function for protein structure prediction

Abstract One of the key elements in protein structure prediction is the ability to distinguish between good and bad candidate structures.
This distinction is made by estimation of the structure energy. The energy function used in the best state-of-the-art automatic
predictors competing in the most recent CASP (Critical Assessment of Techniques for Protein Structure Prediction) experiment
is defined as a weighted sum of a set of energy terms designed by experts. We hypothesised that combining these terms more
freely will improve the prediction quality. To test this hypothesis, we designed a genetic programming algorithm to evolve
the protein energy function. We compared the predictive power of the best evolved function and a linear combination of energy
terms featuring weights optimised by the Nelder–Mead algorithm. The GP based optimisation outperformed the optimised linear
function. We have made the data used in our experiments publicly available in order to encourage others to further investigate
this challenging problem by using GP and other methods, and to attempt to improve on the results presented here.

Content Type Journal Article
Category Original Paper
DOI 10.1007/s10710-009-9087-0
Authors
- Paweł Widera, University of Nottingham School of Computer Science Nottingham NG8 1BB UK
- Jonathan M. Garibaldi, University of Nottingham School of Computer Science Nottingham NG8 1BB UK
- Natalio Krasnogor, University of Nottingham School of Computer Science Nottingham NG8 1BB UK

Journal Genetic Programming and Evolvable Machines
Online ISSN 1573-7632
Print ISSN 1389-2576

Journal Volume Volume 11

Journal Issue Volume 11, Number 1 / March, 2010

Preface

Preface
Content Type Journal ArticleDOI 10.1007/s11047-009-9161-9Authors
Paola Bonizzoni, Milan ItalyGheorghe Paun, Bucharest RomaniaGrzegorz Rozenberg, Leiden The NetherlandsClaudio Zandron, Milan Italy

Journal Natural ComputingOnline ISSN 1…

Preface

Content Type Journal Article
DOI 10.1007/s11047-009-9161-9
Authors
- Paola Bonizzoni, Milan Italy
- Gheorghe Paun, Bucharest Romania
- Grzegorz Rozenberg, Leiden The Netherlands
- Claudio Zandron, Milan Italy

Journal Natural Computing
Online ISSN 1572-9796
Print ISSN 1567-7818

Journal Volume Volume 9

Journal Issue Volume 9, Number 2 / June, 2010

Lecture 1 of the artificial intelligence course

Find below lecture 1 of the course artificial intelligence.

Lecture 1
[Slides – pdf]

Find below lecture 1 of the course artificial intelligence.

Lecture 1

[Slides – pdf]

New issue of SIGEVOlution

The new issue of SIGEVOlution is now available for you to download from:http://www.sigevolution.orgThe issue features:An Interview with Hans-Paul Schwefel with an introduction by Günter RudolphMemetic Algorithms by Natalio KrasnogorLearning From Failu…

The new issue of SIGEVOlution is now available for you to download from:

http://www.sigevolution.org

The issue features:

An Interview with Hans-Paul Schwefel with an introduction by Günter Rudolph
Memetic Algorithms by Natalio Krasnogor
Learning From Failures in Evolutionary Computation @ GECCO-2009
new issues of journals
calls & calendar

The New Issue of SIGEVOlution is Now Available!

The new issue of SIGEVOlution is now available for you to download from:
http://www.sigevolution.org
The issue features:
The issue features:

An Interview with Hans-Paul Schwefel with an introduction by Günter Rudolph
Memetic Algorithms by Natalio Krasnogor
Learning From Failures in Evolutionary Computation @ GECCO-2009
new issues of journals
calls & calendar

The new issue of SIGEVOlution is now available for you to download from:

http://www.sigevolution.org

The issue features:

An Interview with Hans-Paul Schwefel with an introduction by Günter Rudolph
Memetic Algorithms by Natalio Krasnogor
Learning From Failures in Evolutionary Computation @ GECCO-2009
new issues of journals
calls & calendar

Natalio Krasnogor, Steve Gustafson, David A. Pelta, and Jose L. Verdegay (eds): Systems self-assembly: multidisciplinary snapshots

Natalio Krasnogor, Steve Gustafson, David A. Pelta, and Jose L. Verdegay (eds): Systems self-assembly: multidisciplinary snapshots
Content Type Journal ArticleCategory Book ReviewDOI 10.1007/s10710-009-9088-zAuthors
Navneet Bhalla, University of Cal…

Natalio Krasnogor, Steve Gustafson, David A. Pelta, and Jose L. Verdegay (eds): Systems self-assembly: multidisciplinary snapshots

Content Type Journal Article
Category Book Review
DOI 10.1007/s10710-009-9088-z
Authors
- Navneet Bhalla, University of Calgary Department of Computer Science Calgary AB Canada

Journal Genetic Programming and Evolvable Machines
Online ISSN 1573-7632
Print ISSN 1389-2576

Journal Volume Volume 10

Journal Issue Volume 10, Number 4 / December, 2009

All-optical binary flip-flop with the help of Terahertz Optical Asymmetric Demultiplexer

Abstract The memory device is very important as they store various values either temporary or permanently. Optical flip-flop memories
form a fundamental building block for all-optical packet switches in the next generation communication networks. All-optical
flip-flop memory with the help of Terahertz Optical Asymmetric Demultiplexer (TOAD) is proposed and described. Principles
and possibilities of all-optical circuits for TOAD based S–R, J–K, D and T flip-flop are reported. Numerical simulation confirming
described method is also given in this paper.

Content Type Journal Article
DOI 10.1007/s11047-009-9162-8
Authors
- Goutam Kumar Maity, Calcutta Institute of Technology Uluberia, Howrah West Bengal India
- Tanay Chattopadhyay, College of Engineering and Management Department of Physics Kolaghat KTPP Township, Midnapur (East) 721171 West Bengal India
- Dilip Kumar Gayen, College of Engineering and Management Department of Computer Science Kolaghat KTPP Township, Midnapur (East) 721171 West Bengal India
- Chinmoy Taraphdar, Bankura Christian College Department of Physics Bankura West Bengal India
- Anup Kumar Maiti, College of Engineering and Management Department of Physics Kolaghat KTPP Township, Midnapur (East) 721171 West Bengal India
- Santi Prasad Maity, Bengal Engineering College and Science University Department of Information Technology Shibpur, Howrah West Bengal India
- Jitendra Nath Roy, College of Engineering and Management Department of Physics Kolaghat KTPP Township, Midnapur (East) 721171 West Bengal India

Journal Natural Computing
Online ISSN 1572-9796
Print ISSN 1567-7818

Facetwise analysis of XCS for problems with class imbalances

by Albert Orriols-Puig, Ester Bernadó-Mansilla, David E. Goldberg, Kumara Sastry, and Pier Luca Lanzi. IEEE Transactions on Evolutionary Computation, doi=10.1109/ TEVC.2009.2019829, [Publisher site].

Michigan-style learning classifier systems (LCSs) are online machine learning techniques that incrementally evolve distributed subsolutions which individually solve a portion of the problem space. As in many machine learning systems, extracting accurate models from problems with class imbalances—that is, problems in which one of the classes is poorly represented with respect to the other classes—has been identified as a key challenge to LCSs. Empirical studies have shown that Michiganstyle LCSs fail to provide accurate subsolutions that represent the minority class in domains with moderate and large disproportion of examples per class; however, the causes of this failure have not been analyzed in detail. Therefore, the aim of this paper is to carefully examine the effect of class imbalances on different LCS components. The analysis focuses on XCS, which is the most-relevant Michigan-style LCS, although the models could be easily adapted to other LCSs. Design decomposition is used to identify five elements that are crucial to guaranteeing the success of LCSs in domains with class imbalances, and facetwise models that explain these different elements for XCS are developed. All theoretical models are validated with artificial problems. The integration of all these models enables us to identify the sweet spot where XCS is able to scalably and efficiently evolve accurate models of rare classes; furthermore, facetwise analysis is used as a tool for designing a set of configuration guidelines that have to be followed to ensure convergence. When properly configured, XCS is shown to be able to solve highly unbalanced problems that previously eluded solution.