What Are BBNs?
- A Bayesian Belief Network is a graphical representation of probabilistic relationships among different random variables.
- These variables can represent anything: diseases, weather conditions, financial markets, or even whether your cat will knock over that glass of water.
Graphical Structure:
- Imagine nodes (circles) connected by arrows (edges).
- Each node represents a random variable (e.g., “Alarm ringing,” “Burglary,” “Fire”).
- The arrows show dependencies between variables.
Conditional Independence:
- BBNs are conditionally independent.
- This means that each node’s probability depends only on its parents (directly connected nodes).
Example: Alarm System
Let’s consider a simple example: an alarm system in a house.
Nodes:
- Alarm (A): Represents whether the alarm rings.
- Burglary (B): Whether a burglary occurred.
- Fire (F): Whether there’s a fire.
- Person 1 (P1): Whether person 1 calls (hearing the alarm).
- Person 2 (P2): Whether person 2 calls (also hearing the alarm).
Probabilities:
We have observed probabilities for each event:
- P(B=T) = 0.001 (Burglary occurred)
- P(F=T) = 0.002 (Fire occurred)
- P(A|B,T) = 0.95 (Alarm rings given burglary and fire
- P(A|B,T) = 0.95 (Alarm rings given burglary and fire)
- P(A|B,F) = 0.94 (Alarm rings given burglary, no fire)
- P(A|~B,~F) = 0.001 (Alarm rings when no burglary or fire)
Person Nodes:
P1 and P2 call based on the alarm:
- P(P1|A,T) = 0.95 (Person 1 calls if alarm rings)
- P(P2|A,T) = 0.80 (Person 2 calls if alarm rings)
Question:
- What’s the probability that both P1 and P2 call when the alarm rings, but there’s no burglary (B) and no fire (F)?
Solution:
- We compute: P(P1, P2, A, ~B, ~F)
- Using the observed probabilities, we find the joint probability.
Conclusion:
Remember, Baysian belief network's help us reason about complex systems by breaking them down into simpler parts.
Leave Comment