<?xml version='1.0' encoding='UTF-8'?><?xml-stylesheet href='static/style.xsl' type='text/xsl'?><OAI-PMH xmlns="http://www.openarchives.org/OAI/2.0/" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://www.openarchives.org/OAI/2.0/ http://www.openarchives.org/OAI/2.0/OAI-PMH.xsd"><responseDate>2026-11-22T10:01:59Z</responseDate><request verb="GetRecord" identifier="oai:ecommons.cornell.edu:1813/11171" metadataPrefix="dim">https://ecommons.cornell.edu/server/oai/request</request><GetRecord><record><header><identifier>oai:ecommons.cornell.edu:1813/11171</identifier><datestamp>2026-05-14T13:50:02Z</datestamp><setSpec>com_1813_35</setSpec><setSpec>col_1813_47</setSpec></header><metadata><dim:dim xmlns:dim="http://www.dspace.org/xmlns/dspace/dim" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns:doc="http://www.lyncode.com/xoai" xsi:schemaLocation="http://www.dspace.org/xmlns/dspace/dim http://www.dspace.org/schema/dim.xsd">
   <dim:field mdschema="dc" element="contributor" qualifier="author">Niculescu-Mizil, Alexandru</dim:field>
   <dim:field mdschema="dc" element="date" qualifier="accessioned">2008-07-30T01:01:33Z</dim:field>
   <dim:field mdschema="dc" element="date" qualifier="available">2013-07-30T06:24:38Z</dim:field>
   <dim:field mdschema="dc" element="date" qualifier="issued">2008-07-30T01:01:33Z</dim:field>
   <dim:field mdschema="dc" element="identifier" qualifier="uri">https://hdl.handle.net/1813/11171</dim:field>
   <dim:field mdschema="dc" element="identifier" qualifier="bibid">6397233</dim:field>
   <dim:field mdschema="dc" element="description" qualifier="abstract" lang="en_US">First, I consider the problem of simultaneously learning the&#xd;
structures of multiple Bayesian networks from multiple related&#xd;
datasets. I present a multitask Bayes net structure learning&#xd;
algorithm that is able to learn more accurate network structures by&#xd;
transferring useful information between the datasets.  The algorithm&#xd;
extends the score and search techniques used in traditional structure&#xd;
learning to the multitask case by defining a scoring function for&#xd;
sets of structures (one structure for each task) and an&#xd;
efficient procedure for searching for a high scoring set of&#xd;
structures.  I also address the task selection problem in the context&#xd;
of multitask Bayes net structure learning. Unlike in other multitask&#xd;
learning scenarios, in the Bayes net structure learning setting there&#xd;
is a clear definition of task relatedness: two tasks are related if&#xd;
they have similar structures. This allows one to automatically select&#xd;
a set of related tasks to be used by multitask structure learning.&#xd;
&#xd;
&#xd;
Second, I examine the relationship between the predictions made by&#xd;
different supervised learning algorithms and true posterior&#xd;
probabilities. I show that quasi-maximum margin methods such as&#xd;
boosted decision trees and SVMs push probability mass away from 0 and&#xd;
1 yielding a characteristic sigmoid shaped distortion in the predicted&#xd;
probabilities. Naive Bayes pushes probabilities toward 0 and 1. Other&#xd;
models such as neural nets, logistic regression and bagged trees&#xd;
usually do not have these biases and predict well calibrated&#xd;
probabilities.  I experiment with two ways of correcting the biased&#xd;
probabilities predicted by some learning methods: Platt Scaling and&#xd;
Isotonic Regression. I qualitatively examine what distortions these&#xd;
calibration methods are suitable for and quantitatively examine how&#xd;
much data they need to be effective.&#xd;
&#xd;
Third, I present a method for constructing ensembles from libraries of&#xd;
thousands of models.  Model libraries are generated using different&#xd;
learning algorithms and parameter settings.  Forward stepwise&#xd;
selection is used to add to the ensemble the models that maximize its&#xd;
performance.  The main drawback of ensemble selection is that it&#xd;
builds models that are very large and slow at test time. This&#xd;
drawback, however, can be overcome with little or no loss in&#xd;
performance by using model compression.</dim:field>
   <dim:field mdschema="dc" element="description" qualifier="sponsorship" lang="en_US">The work in this dissertation was supported by NSF grants 0347318, 0412930, 0427914, and 0612031.</dim:field>
   <dim:field mdschema="dc" element="language" qualifier="iso" lang="en_US">en_US</dim:field>
   <dim:field mdschema="dc" element="subject" lang="en_US">transfer learning</dim:field>
   <dim:field mdschema="dc" element="subject" lang="en_US">Bayesian network structure learning</dim:field>
   <dim:field mdschema="dc" element="subject" lang="en_US">probability calibration</dim:field>
   <dim:field mdschema="dc" element="subject" lang="en_US">ensemble learning</dim:field>
   <dim:field mdschema="dc" element="title" lang="en_US">Getting the Most Out of Your Data: Multitask Bayesian Network Structure Learning, Predicting Good Probabilities and Ensemble Selection</dim:field>
   <dim:field mdschema="dc" element="type" lang="en_US">dissertation or thesis</dim:field>
   <dim:field mdschema="dc" element="format" qualifier="mimetype">application/pdf</dim:field>
   <dim:field mdschema="dspace" element="entity" qualifier="type">Publication</dim:field>
   <dim:field mdschema="cris" element="virtual" qualifier="collection" authority="https://cornell-ecommons.eks.prod.4science.cloud/handle/1813/47" confidence="600">Cornell Theses and Dissertations</dim:field>
   <dim:field mdschema="cris" element="virtual" qualifier="author">Niculescu-Mizil, Alexandru</dim:field>
   <dim:field mdschema="cris" element="virtualsource" qualifier="collection">5893a6ea-7af3-41d7-abc6-04bcd26ab5df</dim:field>
   <dim:field mdschema="others" element="access-status">open.access</dim:field>
   <dim:field mdschema="others" element="access-status">open.access</dim:field>
   <dim:field mdschema="cerif" element="openaire" authority="" confidence="-1">&lt;Publication xmlns="https://www.openaire.eu/cerif-profile/1.1/" id="58d278aa-dd08-4825-9271-d174fb50f5d1">
	&lt;Type xmlns="https://www.openaire.eu/cerif-profile/vocab/COAR_Publication_Types">http://purl.org/coar/resource_type/c_1843&lt;/Type>
	&lt;Language>en_US&lt;/Language>
   	&lt;Title>Getting the Most Out of Your Data: Multitask Bayesian Network Structure Learning, Predicting Good Probabilities and Ensemble Selection&lt;/Title>
   	&lt;PublishedIn>
    	&lt;Publication>
      	&lt;/Publication>
   	&lt;/PublishedIn>
   	&lt;PublicationDate>2008-07-30T01:01:33Z&lt;/PublicationDate>
   	&lt;Authors>
      	&lt;Author>
        	&lt;DisplayName>Niculescu-Mizil, Alexandru&lt;/DisplayName>
         	&lt;Affiliation>
         		&lt;OrgUnit>
         		&lt;/OrgUnit>
         	&lt;/Affiliation>
      	&lt;/Author>
	&lt;/Authors>
   	&lt;Editors>
	&lt;/Editors>
    &lt;Publishers>
        &lt;Publisher>
            &lt;OrgUnit />
        &lt;/Publisher>
    &lt;/Publishers>
    &lt;Keyword>transfer learning&lt;/Keyword>
    &lt;Keyword>Bayesian network structure learning&lt;/Keyword>
    &lt;Keyword>probability calibration&lt;/Keyword>
    &lt;Keyword>ensemble learning&lt;/Keyword>
   	&lt;Abstract>First, I consider the problem of simultaneously learning the&#xd;
structures of multiple Bayesian networks from multiple related&#xd;
datasets. I present a multitask Bayes net structure learning&#xd;
algorithm that is able to learn more accurate network structures by&#xd;
transferring useful information between the datasets.  The algorithm&#xd;
extends the score and search techniques used in traditional structure&#xd;
learning to the multitask case by defining a scoring function for&#xd;
sets of structures (one structure for each task) and an&#xd;
efficient procedure for searching for a high scoring set of&#xd;
structures.  I also address the task selection problem in the context&#xd;
of multitask Bayes net structure learning. Unlike in other multitask&#xd;
learning scenarios, in the Bayes net structure learning setting there&#xd;
is a clear definition of task relatedness: two tasks are related if&#xd;
they have similar structures. This allows one to automatically select&#xd;
a set of related tasks to be used by multitask structure learning.&#xd;
&#xd;
&#xd;
Second, I examine the relationship between the predictions made by&#xd;
different supervised learning algorithms and true posterior&#xd;
probabilities. I show that quasi-maximum margin methods such as&#xd;
boosted decision trees and SVMs push probability mass away from 0 and&#xd;
1 yielding a characteristic sigmoid shaped distortion in the predicted&#xd;
probabilities. Naive Bayes pushes probabilities toward 0 and 1. Other&#xd;
models such as neural nets, logistic regression and bagged trees&#xd;
usually do not have these biases and predict well calibrated&#xd;
probabilities.  I experiment with two ways of correcting the biased&#xd;
probabilities predicted by some learning methods: Platt Scaling and&#xd;
Isotonic Regression. I qualitatively examine what distortions these&#xd;
calibration methods are suitable for and quantitatively examine how&#xd;
much data they need to be effective.&#xd;
&#xd;
Third, I present a method for constructing ensembles from libraries of&#xd;
thousands of models.  Model libraries are generated using different&#xd;
learning algorithms and parameter settings.  Forward stepwise&#xd;
selection is used to add to the ensemble the models that maximize its&#xd;
performance.  The main drawback of ensemble selection is that it&#xd;
builds models that are very large and slow at test time. This&#xd;
drawback, however, can be overcome with little or no loss in&#xd;
performance by using model compression.&lt;/Abstract>
	&lt;Access xmlns="http://purl.org/coar/access_right" 
    >
    &lt;/Access>
&lt;/Publication>
</dim:field>
</dim:dim>
</metadata></record></GetRecord></OAI-PMH>