Uncertainty | Sahithyan's Notes

In real life systems, AI has to support uncertainty. Probability theory provides a mathematical framework for reasoning under uncertainty.

Required in environments when they are:

Assumes perfect information and predictable outcomes.

Not suitable in uncertain domains, because of:

Assumes that outcomes have likelihoods.

Probability captures degrees of belief, summarizing uncertainty due to ignorance.

Uncertainty can be approached with either:

Non-monotonic logic
Retractable conclusions (e.g., “Birds can fly” except penguins).
Probability theory
Quantitative model of uncertainty.

Probability Theory

Revise Probability basics covered in semester 2.

$\Omega$ is the set of all possible world descriptions. Elements of $\Omega$ are atomic (cannot be broken down).

Aka. probability space. Defined by sample space $\Omega$ and probability function $P$ which assigns a probability for all elements of $\Omega$ .

Given a joint distribution, probability of any proposition φ:

P(φ)=\sum_{\omega: \omega\models φ} P(ω)

Conditional inference uses normalization:

P(\text{Cavity}|\text{Toothache})=\alpha P(\text{Cavity},\text{Toothache})

Here $\alpha$ is a constant ensuring probabilities sum to 1.

In general:

P(Y|E=e)=α\sum_h P(Y,E=e,H=h)

Here $H$ means hidden variables. Time complexity is $O(2^n)$ for $n$ binary variables.

$A$ and $B$ are conditionally independent given $C$ if:

P(A,B|C)=P(A|C)P(B|C)

Reduces required parameters from exponential to linear in n. Useful for efficient probabilistic reasoning.

Assumes conditional independence of effects given the cause:

P(\text{Cause},\text{Effect}_1,…,\text{Effect}_n)=P(\text{Cause})\prod_i P\left(\text{Effect}_i|\text{Cause}\right)

Simplifies computation—parameters grow linearly with n. Example: P(Cavity|toothache ∧ catch) computed from individual conditional probabilities.