Learning With Complete Data
Download as PPTX, PDF2 likes8,112 views
This document discusses different approaches for learning with complete data, including: 1) Parameter learning aims to find numerical parameters for a fixed probability model given complete data for all variables. 2) Maximum likelihood parameter learning derives parameter expressions as log terms and finds values by equating logs to 0. 3) Naive Bayes models assume attributes are conditionally independent, and truth is not representable as a decision tree. 4) Continuous models represent real-world applications using linear Gaussian models that minimize sum of squared errors via standard linear recursion.
Learning With Complete Data
- 1. Learning with Complete Data BY Saravana Kumar.M Vishnuprabhu.G
- 2. Learning with Complete Data • Parameter learning • Maximum-likelihood parameter learning • Naïve Bayes models • Continuous Model • Bayesian Parameter Learning
- 3. Parameter learning • To find the numerical parameters for a probability model whose structure is fixed • Data are complete when each data point contains values for every variable in the model
- 4. Maximum-likelihood parameter learning • discrete model • 3 Steps: – Write expression as finite of parameter – Derive those expressions as log terms – Find parameter value by equating log terms to 0
- 6. (From the text book web site) CSE 471/598 by H. Liu 6
- 7. (From the text book web site) CSE 471/598 by H. Liu 7
- 8. Naïve Bayes Model • Attributes are conditionally independent to each other • Truth of hypothesis is not representable as a decision tree
- 9. Continuous Model • To represent real world applications • Example: linear Gaussian model(similar to discrete data generation) • This model contains sum of squared errors • It can be minimized by standard linear recursion • By minimizing this error gives Maximum- likelihood model
- 10. Bayesian Parameter Learning • It places an hypothesis priority over the possible values of parameter. • It uses Beta Distribution(Conjugate priority)
- 11. Parameter Independence • P(Ѳ1,Ѳ2,Ѳ3) can be represented as p(Ѳ1) p(Ѳ2) p(Ѳ3). LEARNING BASED NET STRUCTURE • This approach is to search for a good model. • It over comes all the disadvantages of all the above models
- 12. • In AI 17% of Learning Theory is based on learning with complete data (LWCD)..!
- 13. THANK YOU