Probabilistic Modelling and Reasoning

— Introduction —

Michael Gutmann

Probabilistic Modelling and Reasoning (INFR11134)

School of Informatics, University of Edinburgh

Spring Semester 2020

Variability

I Variability is part of nature

I Human heights vary
I Men are typically taller than women but
height varies a lot

Michael Gutmann PMR Introduction 2 / 23

Variability

I Our handwriting is unique

I Variability leads to uncertainty: e.g. 1 vs 7 or 4 vs 9

Michael Gutmann PMR Introduction 3 / 23

Variability

I Variability leads to uncertainty

I Reading handwritten text in a
foreign language

Michael Gutmann PMR Introduction 4 / 23

Example: Screening and diagnostic tests

I Early warning test for Alzheimer’s disease (Scharre, 2010, 2014)

I Detects “mild cognitive impairment”

I Takes 10–15 minutes

I Freely available
I Assume a 70 year old man
tests positive.
I Should he be concerned?

(Example from sagetest.osu.edu)

Michael Gutmann PMR Introduction 5 / 23

Accuracy of the test

I Sensitivity of 0.8 and specificity of 0.95 (Scharre, 2010)

I 80% correct for people with impairment

impairment
detected (y=1)

with impairment (x=1)

0.8

0.2

no impairment
detected (y=0)

Michael Gutmann PMR Introduction 6 / 23

Accuracy of the test

I Sensitivity of 0.8 and specificity of 0.95 (Scharre, 2010)

I 95% correct for people w/o impairment

impairment
detected (y=1)

w/o impairment (x=0)

5
0.0

0.9
5

no impairment
detected (y=0)

Michael Gutmann PMR Introduction 7 / 23

Variability implies uncertainty

I People of the same group do not have the same test results
I Test outcome is subject to variability
I The data are noisy
I Variability leads to uncertainty
I Positive test ≡ true positive ?
I Positive test ≡ false positive ?
I What can we safely conclude from a positive test result?
I How should we analyse such kind of ambiguous data?

Michael Gutmann PMR Introduction 8 / 23

Probabilistic approach

I The test outcomes y can be described with probabilities

sensitivity = 0.8 ⇔ P(y = 1|x = 1) = 0.8

⇔ P(y = 0|x = 1) = 0.2
specificity = 0.95 ⇔ P(y = 0|x = 0) = 0.95
⇔ P(y = 1|x = 0) = 0.05

I P(y |x ): model of the test specified in terms of (conditional)

probabilities
I x ∈ {0, 1}: quantity of interest (cognitive impairment or not)

Michael Gutmann PMR Introduction 9 / 23

Prior information
Among people like the patient, P(x = 1) = 5/45 ≈ 11% have a
cognitive impairment (plausible range: 3% – 22%, Geda, 2014)

With impairment
p(x=1)

Without impairment
p(x=0)

Michael Gutmann PMR Introduction 10 / 23

Probabilistic model

I Reality:
I properties/characteristics of the group of people like the
patient
I properties/characteristics of the test
I Probabilistic model:
I P(x = 1)
I P(y = 1|x = 1) or P(y = 0|x = 1)
P(y = 1|x = 0) or P(y = 0|x = 0)
Fully specified by three numbers.
I A probabilistic model is an abstraction of reality that uses
probability theory to quantify the chance of uncertain events.

Michael Gutmann PMR Introduction 11 / 23

If we tested the whole population

With impairment
p(x=1)

Without impairment
p(x=0)
Michael Gutmann PMR Introduction 12 / 23
If we tested the whole population
Fraction of people who are impaired and have positive tests:
P(x = 1, y = 1) = P(y = 1|x = 1)P(x = 1) = 4/45 (product rule)

With impairment
p(x=1)

Without impairment
p(x=0)
Michael Gutmann PMR Introduction 13 / 23
If we tested the whole population
Fraction of people who are not impaired but have positive tests:
P(x = 0, y = 1) = P(y = 1|x = 0)P(x = 0) = 2/45 (product rule)

With impairment
p(x=1)

Without impairment
p(x=0)
Michael Gutmann PMR Introduction 14 / 23
If we tested the whole population
Fraction of people where the test is positive:
P(y = 1) = P(x = 1, y = 1) + P(x = 0, y = 1) = 6/45 (sum rule)

With impairment
p(x=1)

Without impairment
p(x=0)
Michael Gutmann PMR Introduction 15 / 23
Putting everything together

I Among those with a positive test, fraction with impairment:

P(y = 1|x = 1)P(x = 1) 4 2

P(x = 1|y = 1) = = =
P(y = 1) 6 3
I Fraction without impairment:

P(y = 1|x = 0)P(x = 0) 2 1

P(x = 0|y = 1) = = =
P(y = 1) 6 3
I Equations are examples of “Bayes’ rule”.
I Positive test increased probability of cognitive impairment
from 11% (prior belief) to 67%, or from 6% to 51%.
I 51% ≈ coin flip

Michael Gutmann PMR Introduction 16 / 23

Probabilistic reasoning

I Probabilistic reasoning ≡ probabilistic inference:

Computing the probability of an event that we have not or
cannot observe from an event that we can observe
I Unobserved/uncertain event, e.g. cognitive impairment x = 1
I Observed event ≡ evidence ≡ data, e.g. test result y = 1
I “The prior”: probability for the uncertain event before having
seen evidence, e.g. P(x = 1)
I “The posterior”: probability for the uncertain event after
having seen evidence, e.g. P(x = 1|y = 1)
I The posterior is computed from the prior and the evidence via
Bayes’ rule.

Michael Gutmann PMR Introduction 17 / 23

Key rules of probability
(1) Product rule:

P(x = 1, y = 1) = P(y = 1|x = 1)P(x = 1)

= P(x = 1|y = 1)P(y = 1)

(2) Sum rule:

P(y = 1) = P(x = 1, y = 1) + P(x = 0, y = 1)

Bayes’ rule (conditioning) as consequence of the product rule

P(x = 1, y = 1) P(y = 1|x = 1)P(x = 1)

P(x = 1|y = 1) = =
P(y = 1) P(y = 1)
Denominator from sum rule, or sum rule and product rule
P(y = 1) = P(y = 1|x = 1)P(x = 1) + P(y = 1|x = 0)P(x = 0)

Michael Gutmann PMR Introduction 18 / 23

Key rules or probability

I The rules generalise to the case of multivariate random

variables (discrete or continuous)
I Consider the conditional joint probability density function
(pdf) or probability mass function (pmf) of x, y: p(x, y)
(1) Product rule:

p(x, y) = p(x|y)p(y)
= p(y|x)p(x)

(2) Sum rule:

(P
x p(x, y) for discrete r.v.
p(y) = R
p(x, y)dx for continuous r.v.

Michael Gutmann PMR Introduction 19 / 23

Probabilistic modelling and reasoning

I Probabilistic modelling:
I Identify the quantities that relate to the aspects of reality that
you wish to capture with your model.
I Consider them to be random variables, e.g. x, y, z, with a joint
pdf (pmf) p(x, y, z).
I Probabilistic reasoning:
I Assume you know that y ∈ E (measurement, evidence)
I Probabilistic reasoning about x then consists in computing

p(x|y ∈ E)

or related quantities like argmaxx p(x|y ∈ E) or posterior

expectations of some function g of x, e.g.
Z
E [g(x) | y ∈ E] = g(u)p(u|y ∈ E)du

Michael Gutmann PMR Introduction 20 / 23

Solution via product and sum rule

Assume that all variables are discrete valued, that E = {yo }, and
that we know p(x, y, z). We would like to know p(x|yo ).
p(x,yo )
I Product rule: p(x|yo ) = p(yo )
P
I Sum rule: p(x, yo ) = z p(x, yo , z)
P P
I Sum rule: p(yo ) = x p(x, yo ) = x,z p(x, yo , z)
I Result: P
z p(x, yo , z)
p(x|yo ) = P
x,z p(x, yo , z)

Michael Gutmann PMR Introduction 21 / 23

What we do in PMR

P
p(x,yo ,z)
p(x|yo ) = P z p(x,y ,z)
o
x,z

Assume that x, y, z each are d = 500 dimensional, and that each

element of the vectors can take K = 10 values.

I Issue 1: To specify p(x, y, z), we need to specify

K 3d − 1 = 101500 − 1 non-negative numbers, which is
impossible.
Topic 1: Representation What reasonably weak assumptions
can we make to efficiently represent p(x, y, z)?

Michael Gutmann PMR Introduction 22 / 23

What we do in PMR
P
p(x,y ,z)
p(x|yo ) = Pz o
p(x,yo ,z)
x,z

I Issue 2: The sum in the numerator goes over the order of

K d = 10500 non-negative numbers and the sum in the
denominator over the order of K 2d = 101000 , which is
impossible to compute.
Topic 2: Exact inference Can we further exploit the
assumptions on p(x, y, z) to efficiently compute the posterior
probability or derived quantities?
I Issue 3: Where do the non-negative numbers p(x, y, z) come
from?
Topic 3: Learning How can we learn the numbers from data?
I Issue 4: For some models, exact inference and learning is too
costly even after fully exploiting the assumptions made.
Topic 4: Approximate inference and learning
Michael Gutmann PMR Introduction 23 / 23