Scribd
Upload
6 views

Lecture4 DONT USE

This document discusses Bayesian belief networks, which are probabilistic graphical models that represent conditional dependencies between variables. The key points are: - Bayesian belief networks use directed acyclic graphs to represent relationships between variables. - They allow calculating joint and conditional probabilities using Bayes' rule and the Markov assumption that variables are conditionally independent of their non-descendants given their parents. - Learning algorithms can determine the network structure that best fits observed data. - An example network is given to calculate the probability of various events like wet grass given evidence like sprinklers or rain.

Uploaded by

Lamia
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
6 views

Lecture4 DONT USE

This document discusses Bayesian belief networks, which are probabilistic graphical models that represent conditional dependencies between variables. The key points are: - Bayesian belief networks use directed acyclic graphs to represent relationships between variables. - They allow calculating joint and conditional probabilities using Bayes' rule and the Markov assumption that variables are conditionally independent of their non-descendants given their parents. - Learning algorithms can determine the network structure that best fits observed data. - An example network is given to calculate the probability of various events like wet grass given evidence like sprinklers or rain.

Uploaded by

Lamia
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 13

Bayesian Belief

Network
Farhana Shahid, Brac University

Summer 2020
Probabilistic Models

• Naïve Bayes
• Used for classification based on features
• Bayesian Belief Networks
• A compact graphical representation used for causal inference and to represent statistical
relationships
• Probability + Graph Theory
• Bayesian Networks are graphs. In fact, Directed Acyclic Graphs

2
Directed Acyclic Graph

3
Learning the Network

• Given the data, we would like to learn the network that fits
the data well
• Algorithms exist to do this efficiently, the optimal ones are
NP complete
• If many models fit the data well, then we need a
discriminant measure to select
• Scoring function: Bayesian Likelihood

4
Bayes Probability
Rules of Inference
Bayes Logic

• Given our knowledge, that an event may have been the result of two or more causes occurring, what is
the probability it occurred as a result of a particular cause?
• We would like to predict the unobserved, using our knowledge, i.e. assumptions, about things

6
Conditional Probabilities

• •

7
Joint Probabilities

• The probability of all events


• P(AB) = P(A) P(B|A)
• P(ABCD) = P(A) P(B|A) P(C|AB) P(D|ABC)
• For n variables, it takes 2n terms to write it out!

8
Conditional Independencies

• If two events A and B are independent of one another,


• P(AB) = P(A) P(B)
• If event A is independent of B given C then,
• P(A| B, C) = P(A|C) P(B|C)

9
Markov Assumption

10
Example Cloudy
P(C)
T 0.5
F 0.5 Sprinkler Rain

P(S|C)
C P(S=F) P(S=T)
P(R|C) Wet grass
F 0.5 0.5
C P(R=F) P(R=T)
T 0.9 0.1
F 0.8 0.2
T 0.2 0.8

P(W| S, R)
P(W, R, S, C) = P(C) P(S|C) P(R|C, S) P(W| C, R, S)
S R P(W=F) P(W=T)
= P(C) P(S|C) P(R|C) P(W| S, R)
F F 1 0
F T 0.1 0.9
T F 0.1 0.9
T T 0.01 0.99
11
Inference
•

P(W, R, S, C)=
P(C) P(S|C) P(R|C) P(W| S, R)
C R S W
0 0 0.5 × 0.5 × 0.8 × 0.9 = 0.18
0 1 0.5 × 0.5 × 0.2 × 0.99 = 0.0495
1 1
1 0 0.5 × 0.1 × 0.2 × 0.9 = 0.009
1 1 0.5 × 0.1 × 0.8 × 0.99 = 0.0396
∑ = 0.2781 P(W| S, R)
S R P(W=F) P(W=T)
P(R|C) P(S|C) F F 1 0
P(C) C P(S=F) P(S=T) F T 0.1 0.9
C P(R=F) P(R=T)
T 0.5 F 0.5 0.5 T F 0.1 0.9
F 0.8 0.2
F 0.5 T 0.9 0.1 T T 0.01 0.99
T 0.2 0.8
12
DIY

• P(R=1 | W=1) = ?
• Compare P(S=1 | W=1) and P(R=1 | W=1)
• Infer whether the sprinkler was open or it rained in the morning if you find the grass wet in your lawn?

13

You might also like