Artificial Intelligence

Lecture 6：Probabilistic Reasoning

Xiaojin Gong
2021-04-12

Credits: AI Course in Berkeley & JHU

1
Review
 Knowledge and reasoning
 Knowledge base
 Inference
 Propositional logic
 Inference: model checking, resolution, forward chaining

2
Review

3
Outline
 Probability
 Random variables
 Joint and marginal distributions
 Conditional distribution
 Product rule, chain rule, Bayes’ rule
 Inference
 Independence and conditional independence

4
Uncertainty
 Task environments:
 Partially observable
 Non-deterministic

5
Uncertainty
 General situation:
 Observed variables (evidence): Agent knows certain things about
the state of the world (e.g., sensor readings or symptoms)
 Unobserved variables: Agent needs to reason about other aspects
(e.g. where an object is or what disease is present)
 Model: Agent knows something about how the known variables
relate to the unknown variables
 Probabilistic reasoning gives us a framework for managing
our beliefs and knowledge

6
Making Decision Under Uncertainty
 Example:

 Which action to choose?

 Depends on my preferences for missing flight vs. airport cuisine, etc.
 Utility theory is used to represent and infer preferences
 Decision theory = probability theory + utility theory

7
Rational Decision

 Decision theory = probability theory + utility theory

 An agent is rational iff it chooses the action that yields the highest
expected utility, averaged over all the possible outcomes of the
action.
 A.k.a the principle of maximum expected utility.
8
Random Variable
 A random variable is some aspect of the world about which
we have uncertainty, e.g.
 R = Is it raining? {true, false}
 T = Is it hot or cold? {hot, cold}
 D = How long will it take to drive to airport? [0, ∞)

 We denote random variables with capital letters

9
Probability Distribution
 Associate a probability with each value
∀𝑥 𝑃 𝑋 = 𝑥 ≥ 0，and 𝑃 𝑋=𝑥 =1
𝑥

 Temperature  Weather

10
Probability for Discrete Variables
 Unobserved random variables have distributions
 A distribution is a TABLE of probabilities of values
 Temperature  Weather

11
Probability for Continuous Variables
 Probability density function
𝑃 𝑥 ≤ 𝑋 ≤ 𝑥 + 𝑑𝑥
𝑃 𝑥 = lim
𝑑𝑥→0 𝑑𝑥

12
Joint Probability Distribution
 A joint probability distribution over a set of random
variables: 𝑋1 , ⋯ , 𝑋𝑛 specifies a real number of each
assignment (or outcome):
 Denote: 𝑃 𝑋1 = 𝑥1 , ⋯ , 𝑋𝑛 = 𝑥𝑛 = 𝑃 𝑥1 , ⋯ , 𝑥𝑛

𝑃 𝑥1 , ⋯ , 𝑥𝑛 ≥ 0 , 𝑎𝑛𝑑 𝑃 𝑥1 , ⋯ , 𝑥𝑛 = 1
𝑥1 ,⋯,𝑥𝑛

13
Probabilistic Models
 A probabilistic model is a joint distribution over a set of
random variables
 Probabilistic models:
 Random variables with domains
 Assignments are called outcomes
 Joint distributions: say whether assignments (outcomes) are likely
 Normalized: sum to 1.0
 Ideally: only certain variables directly interact

14
Events
 An event is a set E of outcomes
𝑃 𝐸 = 𝑃 𝑥1 , ⋯ , 𝑥𝑛
(𝑥1 ,⋯,𝑥𝑛 )∈𝐸

 From a joint distribution, we can calculate the probability of

any event:
 Probability that it’s hot AND sunny?
 Probability that it’s hot?
 Probability that it’s hot OR sunny?

 Typically, the events we care about are partial assignments,

like P(T=hot)
15
Marginal Distributions
 Marginal distributions are sub-tables which eliminate variables
 Marginalization (summing out): Combine collapsed rows by adding

𝑃 𝑇 = 𝑃(𝑇, 𝑊)
𝑊

𝑃 𝑊 = 𝑃(𝑇, 𝑊)
𝑇

16
Conditional Probabilities
 Conditional distributions are probability distributions over
some variables given fixed values of others
𝑃(𝑊 = 𝑠, 𝑇 = 𝑐) 0.2
𝑃 𝑊=𝑠𝑇=𝑐 = = = 0.4
𝑃(𝑇 = 𝑐) 0.5

 Joint probability:
𝑃 𝑠, 𝑐 = 𝑃 𝑠 𝑐 𝑃(𝑐)

17
Conditional Probability Distribution
 Normalization:
𝑃(𝑊 = 𝑠, 𝑇 = 𝑐)
𝑃 𝑊=𝑠𝑇=𝑐 =
𝑃(𝑇 = 𝑐)
𝑃(𝑇 = 𝑐, 𝑊 = 𝑠)
=
𝑊 𝑃(𝑇 = 𝑐, 𝑊)

18
Independence
 Two variables are independent (𝑋 ⊥ 𝑌) in a joint distribution
if:
𝑃 𝑋, 𝑌 = 𝑃 𝑋 𝑃 𝑌
∀𝑥, 𝑦 𝑃 𝑥, 𝑦 = 𝑃 𝑥 𝑃(𝑦)

19
Conditional Independence
 X is conditionally independent of Y given Z (𝑋 ⊥ 𝑌|𝑍) iff:
∀𝑥, 𝑦, 𝑧 𝑃 𝑥, 𝑦|𝑧 = 𝑃 𝑥|𝑧 𝑃(𝑦|𝑧)

 Conditional independence is our most basic and robust form

of knowledge about uncertain environments.
 Unconditional independence very rare.

20
Probabilistic Inference
 Probabilistic inference: compute a desired probability from
other known probabilities
 We generally compute conditional probabilities
 P(on time | no reported accidents) = 0.9
 These represents the agent’s beliefs given the evidence
 Probabilities change with new evidence:
 P(on time | no accidents, 5 a.m.) = 0.95
 P(on time | no accidents, 5 a.m., raining) = 0.80
 Observing new evidence causes beliefs to be updated

21
Inference by Enumeration
 General case:
 Evidence variables: 𝐸1 ⋯ 𝐸𝑘 = 𝑒1 ⋯ 𝑒𝑘
 Query variable: 𝑄 𝑋1 ⋯ 𝑋𝑛
 Hidden variables: 𝐻1 ⋯ 𝐻𝑘 All variables
 We want: 𝑃(𝑄|𝑒1 ⋯ 𝑒𝑘)
 Solution:
 Step 1: Select the entries consistent with the evidence
 Step 2: Sum out H to get joint of Query and evidence
 Step 3: Normalize

22
Inference by Enumeration
 P(sun)?

 P(sun | winter)?

 P(sun | winter, hot)?

23
Inference by Enumeration

 Worst-case time complexity O(dn)

 Space complexity O(dn) to store the joint distribution

24
The Product Rule
 Infer the joint probability from conditional distributions

𝑃 𝑥, 𝑦 = 𝑃 𝑦 𝑃 𝑥 𝑦

25
The Chain Rule
 The joint distribution can be written as an incremental
product of conditional distributions:

𝑃 𝑥1, 𝑥2, 𝑥3 = 𝑃 𝑥1 𝑃 𝑥2 𝑥1 𝑃 𝑥3 𝑥1, 𝑥2

𝑃 𝑥1, 𝑥2, … , 𝑥𝑛 = 𝑃(𝑥𝑖 |𝑥1 … 𝑥𝑖−1 )

26
Bayes’ Rule
 Two ways to factor a joint distribution:

𝑃 𝑥, 𝑦 = 𝑃 𝑦 𝑃 𝑥 𝑦 = 𝑃 𝑥 𝑃 𝑦 𝑥

𝑃 𝑦𝑥
⇒𝑃 𝑥𝑦 = 𝑃 𝑥
𝑃 𝑦

⇒𝑃 𝑥𝑦 ∝𝑃 𝑦𝑥 𝑃 𝑥

Posterior Likelihood Prior

27
Bayes’ Rule
𝑃 𝑦𝑥
𝑃 𝑥𝑦 = 𝑃 𝑥
𝑃 𝑦

 Why is this at all helpful?

 Build one conditional from its reverse
 Often one conditional is tricky but the other one is simple
 Foundation of many systems
 In the running for most important AI equation!

28
Inference with Bayes’ Rule
 Example:
 Diagnostic probability from causal probability:

 Example:
 Given:

 What is P(W | dry) ?

29
The Wumpus World


30
The Wumpus World
 Specifying the Probability Model

31
The Wumpus World
 Observations and Query

32
The Wumpus World
 Using Conditional Independence

33
The Wumpus World


34
The Wumpus World


35
Assignments
 Reading assignment:
 Ch. 13

36