Week 10 - Sequential Modelling, Hidden Mark

Question 1

Define HMMs

Answer 1

It stands as Hidden Markov models

Question 2

What are two most important concepts in HMMs?

Answer 2

1. Each hidden state Yt ∈ ΩY with K distinct values depends only upon the one before it in time Yt-1 for all t = 0,1,…,T
2. The measured observations Xt depend only upon the associated hidden state, Yt

Question 3

What does HMMs capture?

Answer 3

It capture the time-dependent RVs where it is not directly measured

Question 4

What is the goal that achieved using HMMs?

Answer 4

Given fixed model parameters and observed data X0,X1,…,XT, computer the most probable sequence of hidden states y = [y0, 𝑦1, …, 𝑦*T]

Question 5

What are the examples that is shown in the HMM intuition?

Answer 5

1. Smart-watch based activity monitoring system
2. Measured observation (Xt): heart rate discrete levels (high/low)
3. Hidden state (Yt): activity level (rest/exercise)

Question 6

What are the evaluation presented in the smartwatch-based activity monitoring system?

Answer 6

1. Single evaluation: if the HR tends to be low, what’s the most probable activity level of the user?
2. Sequence evaluation: if two consecutive measurements, we get HR [low,high]

Question 7

Sequence evaluation

Question 8

What is the problem seen in genome sequence segmentation?

Answer 8

That optimal segmentation of DNA sequences into exon (e), intron (i) donor site (d) sub-sequences

Question 9

What are the distributions seen in GSS?

Answer 9

The distribution is: Y – regions, ΩY={e,d,i}, X – DNA sequences, ΩX={a,c,g,t}

Question 10

What about the state transition distribution and observation distributions seen in GSS?