Thanks to visit codestin.com
Credit goes to www.scribd.com

Scribd
Upload
40 views2 pages

E-M Algorithm Questions

The document contains a practice sheet for a machine learning course focusing on the Expectation-Maximization (EM) algorithm, with various questions related to Gaussian and Poisson mixture models, responsibilities, and parameter updates. It includes computations for log-likelihood, posterior probabilities, expected assignment probabilities, and updates for Beta and Gamma distributions. Additionally, it addresses the theoretical aspects of the EM algorithm, such as its steps, convergence, and suitability for estimating parameters in incomplete data scenarios.

Uploaded by

kingslayer
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
40 views2 pages

E-M Algorithm Questions

The document contains a practice sheet for a machine learning course focusing on the Expectation-Maximization (EM) algorithm, with various questions related to Gaussian and Poisson mixture models, responsibilities, and parameter updates. It includes computations for log-likelihood, posterior probabilities, expected assignment probabilities, and updates for Beta and Gamma distributions. Additionally, it addresses the theoretical aspects of the EM algorithm, such as its steps, convergence, and suitability for estimating parameters in incomplete data scenarios.

Uploaded by

kingslayer
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

BITS F464 Machine Learning E-M Algorithm Practice Sheet

Q1. Verstappen’s race in dry conditions is modelled as a mixture of two Gaussians. Given Gaussian
1 𝑥−µ 2
1 −2( )
Distribution: 𝑒 σ

2πσ
2
●​ Quali Mode: µ1 = 87, σ1 = 2, π1 = 0. 7
2
●​ Fuel-Saving Mode: µ2 = 90, σ2 = 2, π2 = 0. 3
Give lap times X = {88, 89, 91}. Compute the log-likelihood.

Q2. Drivers adjust their driving styles based on tyre wear. Given initial means: µ1 = 88 & µ2 = 92,
computed responsibilities: γ(𝑧𝑖1) = {0. 8, 0. 5, 0. 2} & γ(𝑧𝑖2) = {0. 2, 0. 5, 0. 8}, and lap times:
𝑋 = {87, 90, 94}. Update the new mean values using the Maximisation Step.

2
Q3. In a race, Scuderia Ferrari’s lap times follow a Gaussian µ1 = 91 & σ1 = 1 and Red Bull Racing are
2
modelled with µ2 = 88 & σ1 = 1. Given observed lap times X = {89, 92, 87} and π1 = π2 = 0. 5.
Compute the posterior probability that each lap was set by Scuderia Ferrari.

Q4. Torro Rosso pit stop errors are modelled using a Poisson Mixture Model. Given Poisson Distribution:
−λ 𝑥
𝑒 λ
𝑃(𝑋 = 𝑥 | λ) = 𝑥!
. There are two types of errors:
●​ Minor Errors λ1 = 1 per pit stop
●​ Major Errors λ2 = 4 per pit stop
Given pit stop error counts X = {1, 2, 0, 5, 3, 1} with initial π1 = 0. 7 & π2 = 0. 3. Compute expected
assignment probabilities for each pit stop.

Q5. In a wet Monaco Grand Prix, we suspect two types of drivers - Rain Masters and Strugglers. Given
the observed lap time (in s): 𝑋 = {75. 2, 76. 1, 80. 5, 81. 3, 82. 9}. Assume a Gaussian Mixture Model
(GMM) with two components initialised with µ1 = 76, µ2 = 81, σ1 = σ2 = 1 & π1 = π2 = 0. 5 .
Compute initial responsibilities for each lap time.

Q6. Answer briefly: Explain the two main steps of the Expectation-Maximization (EM) algorithm. Why is
the E-step necessary, and what does it compute? What happens in the M-step, and how does it update
parameters?

Q7. F1 teams get known sponsors and unknown investors. We model the funding distribution using the
α
β α−1 −β𝑥
Gamma Mixture Model. Given Gamma Distribution: 𝑓(𝑥, α, β) = Γ(α)
𝑥 𝑒
●​ Sponsor Money: 𝐺𝑎𝑚𝑚𝑎(α1 = 3, β1​ = 1)
●​ Hidden Investors: 𝐺𝑎𝑚𝑚𝑎(α2 = 5, β2​ = 2)
Given observed funding amounts (in million $) X = {3, 4, 6, 8, 2, 7, 5} with π1 = 0. 6 & π2 = 0. 4.
Compute the expected probability that 6 M $ came from hidden investors.

Q8. Explain in short:


1.​ Why does the EM algorithm always increase the likelihood in each iteration?
2.​ Can EM guarantee convergence? If so, what does it converge to? Explain in short.

Q9. A team models the risk of spinning out in wet conditions using Beta distributions. Given Beta
Γ(α+β) α−1 β−1
Distribution: Γ(α)Γ(β)
𝑥 (1 − 𝑥)
●​ Careful Drivers: α1 = 4 & β1 = 6
●​ Aggressive Drivers: α2 = 8 & β2 = 2

Observed spin counts (per 10 laps): X = {1, 2, 3, 0, 2, 1}. Compute the updated Beta parameters after one
E-M step.

Q10. A company has two types of defective machines, Type A and Type B. When a machine fails, the
company records the failure but does not know which type of machine failed.

●​ Type A fails with probability 𝑝𝐴​and Type B fails with probability 𝑝𝐵.
●​ A randomly chosen machine is of Type A with probability λ and of Type B with probability 1−λ.

(a) Why is the Expectation-Maximization (E-M) algorithm suitable for estimating 𝑝𝐴, 𝑝𝐵 and λ given this
incomplete data?

(b) What challenges might arise when using the E-M algorithm for this problem?

You might also like