8.4.1-Learning-Overview

8.4.1-Learning-Overview - Machine Learning ! ! ! ! !...

Info iconThis preview shows pages 1–9. Sign up to view the full content.

View Full Document Right Arrow Icon
Machine Learning Srihari 1 Learning Graphical Models: Overview Sargur Srihari [email protected]
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Machine Learning Srihari Topics • Motivation • Goals of Learning 1. Density Estimation 2. Specific Prediction Tasks 3. Knowledge Discovery • Learning as Optimization Empirical Risk and Overfitting Discriminative vs Generative Training • Learning Tasks Model Constraints Data Observability Taxonomy of Learning Tasks 2 Alice Debbie Bob Charles BN MN
Background image of page 2
Machine Learning Srihari Motivation • Usual starting point is a given graphical model – Structure and parameters are part of input • Two approaches to task of acquiring a model 1. Knowledge Engineering • Construct a network by hand with expert s help 2. Machine Learning • Learn model from a set of instances
Background image of page 3

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Machine Learning Srihari Knowledge Engineering vs ML • Knowledge Engineering Approach – Pick variables, pick structure, pick probabilities – Effort • Simple ones require hours of effort, complex one: months – Significant testing of model by evaluating results of typical queries yield plausible answers • Machine Learning Approach – Instances available from distribution we wish to model – Easier to get large data sets rather than human expertise
Background image of page 4
Machine Learning Srihari Difficulties with Manual Construction • In some domains: – Amount of knowledge required too large – No experts who have sufficient understanding – Cost: expert time is valuable • Properties of distribution change from one site to another • Change over time – Expert cannot redesign every few weeks • Modeling mistakes have serious impact on quality of answers 5
Background image of page 5

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Machine Learning Srihari Advantage of ML approach • We are in the Information Age – Easier to obtain even large amounts of data in electronic form than to obtain human expertise • Example Data – Medical Diagnosis • Patient records – Pedigree Analysis (Genetic Inheritance) • Family trees for disease transmission – Image Segmentation • Set of images segmented by a person 6
Background image of page 6
Machine Learning Srihari Medical Diagnosis Task • Collection of patient records – History: • Age, sex, history, medical complications – Symptoms – Results of tests – Diagnosis – Treatment – Outcome • Task: Use data to model distribution of patients – Pathologist diagnoses disease of lymph nodes (Pathfinder 1992) 7
Background image of page 7

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Machine Learning Srihari • Set of Family trees • Task: Learn distribution – Breast cancer, Blood Type • Three types of CPDs: – Penetrance model: phenotype given genotype • Probability of a phenotype (say patient s blood type, B ) given person’s genotype (ordered pair of parent s blood- type, A,B or O): P(B(c)|G(c)) c=person, p=father , m=mother – Transmission model: genotype passed to child • How often a genotype (locus for disease or blood type) passed from parent to child: P(G(c)|G(p),G(m)):
Background image of page 8
Image of page 9
This is the end of the preview. Sign up to access the rest of the document.

This document was uploaded on 02/25/2012.

Page1 / 29

8.4.1-Learning-Overview - Machine Learning ! ! ! ! !...

This preview shows document pages 1 - 9. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online