Chapter15 1
Chapter15 1
Probabilistic Reasoning
Over Time
Sukarna Barua
Assistant Professor, CSE, BUET
Temporal Probabilistic Models
Static world (as we considered in Bayesian network):
- Random variables have a fixed number of states/values.
- Values of Random variables doesn’t change over time
- What is the probability that given all previous state and evidence values?
- What is the size of CPT whenis large? [exponentially large]
- Not practical from computational perspective
Markov Assumption for Sensor Model
Assumption: Evidence at time is independent of all previous states and events
evidences
- [probability of emitting output from state ]
- Umbrella was observed on first three days and absent on fourth, the most
likely state sequence could be it rained first three days and did not rain
on fourth.
- Speech recognition: What is the sequence of words given a sequence of
sounds?
Filtering
Compute probability distribution of current state given observation sequence
Agent maintains the probability distribution of current state at time step .
As new evidence comes up, agent updates its estimation of current state
probabilities
Filtering
: a is a normalizing constant to make probabilities sum up to
1
Filtering
How to calculate
- Marginalize over
Filtering
Hence,
Day 2:
- Can be calculated as:
Filtering: Example
Probability of rain increases at day 2 from day 1 [why?]
Prediction
Compute probability distribution of a future state: )
Can be computed using filtering:
- First compute [forward algorithm]
- Then compute as:
- Similarly, compute , …,
The more uncertainty in the transition model, the shorter will be the mixing time
and the more future is obscured!
Likelihood of Evidence Sequence
What is the likelihood of evidence sequence
Compute as
-
[Markov assumption]