Bayesian learning and more general topic
February 15, 2010
- pdf v.s. likihood function
- MLE, estimation
- estimation bias
- BI and court case: Howland will Forgery, Sally Clark, Lucia de B.
- Bayesian search. USS Scorpion P(here|nofind)=(P(nofind|here)*P(here)) ../(P(nofind|here)*P(here)+P(nofind|nohere)*P(nohere))
- clearly there is a strong relationship between Kalman filter and HMM, both of which belongs to Dynamic Beyesian Network, here is a tutorial on this topic. Mackey’s toturial on Beyesian network. Murphy K presentation on DBN, and his ’02 thesis. Murphy K’s publication page. Micheal Jordan’s publication and tutorials. Judea Pearl, Bayesian causality (Judea asking very fundamental and interesting questions here) his publication. this explains why RMF is interested by physicist while DMN is interested by AI community. reasoning(prediction, estimation, classification) is causal, describe nature law is non directional(parameter and variable interchangible, observational(algebra explains observation non-experiment v.s. interventional, control experiment, algebra explains observation + causality lead to prediction(?)).P(r|w) v.s. P(r|do(w)). Simpons paradox and solution; reverse regression (gender, salary, qualification). aka adjustment problem, covariate selection problem. wiki on Belief Propagation just learned that Judea Pearl is father of Denial Pearl.
- K. Murphy tutorial on: RL here, Graphical model and BN here
- Naive Bayes classifier (independent feature)
- a unified review of linear gaussian model (FA, PCA, QFA, KF, HMM unified!@_@)
- Information theory, Inference, Learning algorithm, David MacKay’s book; an intro here