Presentation by Andreas Schleicher Tackling the School Absenteeism Crisis 30 ...
08 probabilistic inference over time
1. Bayesian Networks
Unit 8 Probabilistic Inference
over Time
Wang, Yuan-Kai, 王元凱
ykwang@mails.fju.edu.tw
http://www.ykwang.tw
Department of Electrical Engineering, Fu Jen Univ.
輔仁大學電機工程系
2006~2011
Reference this document as:
Wang, Yuan-Kai, “Probabilistic Inference over Time,"
Lecture Notes of Wang, Yuan-Kai, Fu Jen University, Taiwan, 2011.
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
2. Bayesian Networks Unit - Probabilistic Inference over Time p. 2
Goal of This Unit
• Know the uncertainty concept in temporal
models
• Learn four inference types in temporal
models
– Filtering, Prediction, Smoothing,
Most Likely Explanation
• See some temporal models
– HMM, Kalman/Particle filtering
– Dynamic Bayesian networks
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
3. Bayesian Networks Unit - Probabilistic Inference over Time p. 3
Related Units
• Background
– Probabilistic graphical model
– Exact inference in BN
– Approximate inference in BN
• Next units
– HMM
– Kalman filter
– Particle filter
– DBN
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
4. Bayesian Networks Unit - Probabilistic Inference over Time p. 4
Self-Study Reference
• Chapter 15, Sections 15.1-15.2, Artificial
Intelligence - a modern approach, 2nd, by S.
Russel & P. Norvig, Prentice Hall, 2003.
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
5. Bayesian Networks Unit - Probabilistic Inference over Time p. 5
Structure of Related Lecture Notes
Problem Structure Data
Learning
PGM B E
Representation Learning
A
Unit 5 : BN Units 16~ : MLE, EM
Unit 9 : Hybrid BN J M
Units 10~15: Naïve Bayes, MRF,
HMM, DBN,
Kalman filter P(B) Parameter
P(E) Learning
P(A|B,E)
P(J|A)
Query Inference
P(M|A)
Unit 6: Exact inference
Unit 7: Approximate inference
Unit 8: Temporal inference
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
6. Bayesian Networks Unit - Probabilistic Inference over Time p. 6
Contents
1. Time and Uncertainty …………………...... 7
2. Inference in Temporal Models ……...……. 46
3. Various Models .…….................................... 90
4. References …………………………………. 96
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
7. Bayesian Networks Unit - Probabilistic Inference over Time p. 7
1. Time and Uncertainty
• What is probabilistic reasoning over
time
– There are a lot of time-series data
• Ex: Stock data, weather data, radar
signal, ...
– We want to
• Predict its next data
• Recover correct values of its current data
• Recover correct values of its previous data
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
8. Bayesian Networks Unit - Probabilistic Inference over Time p. 8
Example – Stock Data
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
9. Bayesian Networks Unit - Probabilistic Inference over Time p. 9
Example 2 - Visual Tracking
• What is visual tracking
– Continuously detect objects in video
– Time series data
• What kind of objects
– Face,
– Facial features (eye, eyebrow, ...)
– Human body
– Hand
– ...
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
10. Bayesian Networks Unit - Probabilistic Inference over Time p. 10
Why Visual Tracking (1/2)
• A simple idea to detect objects in all frames
of a video
– "Detect object at every frame with the same
detection method
• Disadvantage
– A detection of a frame may be slow
– Detections at all frames become very slow
• So, if you have a very quick detection
method, the simple method is OK?
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
11. Bayesian Networks Unit - Probabilistic Inference over Time p. 11
Why Visual Tracking (2/2)
• A better approach to detect objects in all
frames of a video
– Detect objects at the first frame
– Find objects at succeeding frames with a quick
method
tracking
• Goal of visual tracking
– Fast and accurate detection of objects
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
12. Bayesian Networks Unit - Probabilistic Inference over Time p. 12
Front-View Face Tracking
Single frame detector
Temporal detector
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
13. Bayesian Networks Unit - Probabilistic Inference over Time p. 13
Side-View Face Tracking
without temporal continuity without temporal continuity
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
14. Bayesian Networks Unit - Probabilistic Inference over Time p. 14
Two Kinds of Approaches
• Neighborhood-based
– Search the neighborhood of the object's
location in previous frame
• Prediction-based
– Search the neighborhood of the predicted
location in current frame
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
15. Bayesian Networks Unit - Probabilistic Inference over Time p. 15
Basic Algorithm
• Basic idea of both the two approaches
1.Read first frame
2.Detect moving object O
Obtain Region of Interest (ROI), usually rectangle or
ellipse
3.Read next frame
4.For all possible ROI candidate Oc
a)Compare the similarity between O and Oc
b)If similarity is high, tracking successfully.
Break
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
16. Bayesian Networks Unit - Probabilistic Inference over Time p. 16
Neighborhood-search Tracking
• Basic idea
1. Read first frame
2. Detect face O
Obtain Region of Interest (ROI), usually rectangle
or ellipse
3. Read next frame
4. For all possible ROI candidate Oc
a) Compare the similarity between O and Oc
b) If similarity is high, break
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
17. Bayesian Networks Unit - Probabilistic Inference over Time p. 17
Basic Ideas
Face
Detection
O
First frame
Face
Tracking
O
Oc
Search Region
Next frame
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
18. Bayesian Networks Unit - Probabilistic Inference over Time p. 18
Prediction-based Tracking
• Three steps
– Predict next position of moving objects
with a probabilistic model (parameters)
– Detect new position around the predicted
position
• Prediction error
– Update
• The correct position
• The probabilistic model with the
prediction error
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
19. Bayesian Networks Unit - Probabilistic Inference over Time p. 19
Predict Next Position
P ( zt | xt )
Current frame
Previous frames
Real position : xt Predicted position
Detected position : zt x-t+1
P ( zt | xt ) P(x t 1 | xt )
Probabilistic
P ( x t | x t 1 ) model
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
20. Bayesian Networks Unit - Probabilistic Inference over Time p. 20
Detect New Position by LSE
Predicted position
Search region
SE = 1032, 2560, LSE = 104
1968, 104, 2223,
... Detected position: zt+1
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
21. Bayesian Networks Unit - Probabilistic Inference over Time p. 21
Update
x-t+1 Prediction Error zt+1
x-t+1-zt+1
Corrected position xt+1
Corrected P '( z t | x t )
Probabilistic
model P '( x t | x t 1 )
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
22. Bayesian Networks Unit - Probabilistic Inference over Time p. 22
Accurate Tracking = Smoothing
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
23. Bayesian Networks Unit - Probabilistic Inference over Time p. 23
Example 3 - Robot Localization
• Localization of AIBO robot in
RoboCup
• The robot has to
– See landmark
• Object detection & object recognition
– Analyze the landmark
• Calculate distance & angle between the
robot and the landmark
– Estimate its location
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
24. Bayesian Networks Unit - Probabilistic Inference over Time p. 24
RoboCup Field
(r , )
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
25. Bayesian Networks Unit - Probabilistic Inference over Time p. 25
Tracking of Robot
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
26. Bayesian Networks Unit - Probabilistic Inference over Time p. 26
Temporal Patterns
• Deterministic patterns :
– Traffic light
– FSM
(Finite State Machine)
–…
• Non-Deterministic patterns :
– Weather
– Speech
– Tracking
– …
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
27. Bayesian Networks Unit - Probabilistic Inference over Time p. 27
How to Do It?
• What we want?
– Prediction: Predict its next data
– Filtering: Recover correct values of its
current data
– Smoothing: Recover correct values of
its previous data
• How to achieve it?
– Statistically model the data
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
28. Bayesian Networks Unit - Probabilistic Inference over Time p. 28
Statistically Modeling
y
x Predict
A set of time-
y = 1.3x + 96 : Model Filter
related data Smooth
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
29. Bayesian Networks Unit - Probabilistic Inference over Time p. 29
State
• There is a set of time-related data
Time t = 0 1 2 3 ...
︵ ︵ ︵ ︵
50, 100 49, 98 50, 96 48, 94
State s = 50, 180 50, 178 50, 176 47, 173
50, 160 49, 158 50, 156 48, 154
︶ ︶ ︶ ︶
• We call each data
– A state of the system, or
– A state of the object
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
30. Bayesian Networks Unit - Probabilistic Inference over Time p. 30
Observable v.s. Unobservable States
• Observable state
– Measurable values
• Sensor values, feature values
– Ex : Localization/Visual Tracking
• Measured position, Measured speed
– Ex : Facial Expression Recognition
• Eyebrow up, eyebrow down, ...
• Unobservable state
– Real state of the system/object
– Ex : Localization/Visual Tracking
• Real position, real speed
– Ex : Facial Expression Recognition
• Smile, Cry, Anger, ...
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
31. Bayesian Networks Unit - Probabilistic Inference over Time p. 31
Observable v.s. Unobservable States
(Math)
• Let
– Xt = set of unobservable state variables at
time t
– Et = set of observable state variables at
time t
• Usually we observe
– E0, E1, ...., Et : time-related data
• But we want to derive
– X0, X1, ..., Xt
• Notation: Xa:b = Xa, Xa+1, ..., Xb
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
32. Bayesian Networks Unit - Probabilistic Inference over Time p. 32
Markov Chain
• Markov chain is an assumption
– A state is dependent on previous state
– Xt depends on X0:t-1
– Xt+1 will not influence Xt
• Markov process
– If we assume that a set of data obeys
Markov assumption,
– We say the data perform Markov
process
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
33. Bayesian Networks Unit - Probabilistic Inference over Time p. 33
Markov Process
• First-order Markov process
– P(Xt |X0:t-1)=P(Xt | Xt-1 )
• Second-order Markov process
– P(Xt |X0:t-1)=P(Xt | Xt-2 , Xt-1 )
• Higher order Markov process ...
– Complicate, seldom used
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
34. Bayesian Networks Unit - Probabilistic Inference over Time p. 34
Transition Model & Sensor Model
• Transition model
– P(Xt | Xt-1 )
– P(Xt | Xt-2 , Xt-1 )
• Sensor model
– We usually assume the evidence
variables (sensor values) at time t, Et,
depend only on the current state Xt
– P(Et|X0:t, E0:t-1) = P(Et|Xt)
– It is also called observation model
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
35. Bayesian Networks Unit - Probabilistic Inference over Time p. 35
Diagram of Transition & Sensor
Models for 1st Order Markov
• P(Xt | Xt-1 ) Transition of
Xt-1 Xt unobservable states
Causal relationship
• P(Et|Xt) Xt Et between observable &
unobservable states
Xt-1 Xt Xt+1 Xt+2
Et Et+1 Et+2
A special Bayesian network
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
36. Bayesian Networks Unit - Probabilistic Inference over Time p. 36
An "Umbrella World" Example (1/2)
• A security guard is always at a secret
underground room, without going out
• He wants to know if it is raining today
• But he can not observe the outside world
• He can only see each morning the
director coming in with, or without, an
umbrella
• Rain is the unobservable state
• Umbrella is the observable state
(sensors values)
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
37. Bayesian Networks Unit - Probabilistic Inference over Time p. 37
An "Umbrella World" Example (2/2)
• For each day t, the set Et contains a single
evidence Ut (whether the umbrella appears)
• The set Xt contains a single state variable Rt
(whether it is raining)
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
38. Bayesian Networks Unit - Probabilistic Inference over Time p. 38
Stationary Process
• The transition model P(Xt | Xt-1) and the
sensor model P(Et | Xt) are all fixed for
all time t
• Stationary process assumption
– Can reduce the complexity of the
algorithm for inference
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
39. Bayesian Networks Unit - Probabilistic Inference over Time p. 39
Inference for the Markov Process (1/2)
X0 X1 X2 Xt
E1 E2 Et
• A Bayesian net with 2 random variables
– X: X0, X1, ..., Xt
– E: E1, ..., Et
• We know that P(X0, X1, ..., Xt, E1, ..., Et),
the FJD, can answer any query
– But it can be reduced t
P( X 0 , X 1 , X t , E1 , , Et ) P( X 0 ) P( X i | X i 1 ) P( Ei | X i )
i 1
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
40. Bayesian Networks Unit - Probabilistic Inference over Time p. 40
Inference for the Markov Process (2/2)
• We need three PDFs
– P(X0), P(Xi|Xi-1), P(Ei|Xi)
• For discrete R.V., we need
– 1 prior probability table P(X0)
– 2 CPTs
• CPT for transition model: P(Xi|Xi-1)
• CPT for sensor model: P(Ei|Xi)
• For continuous R.V., we need
– Gaussian pdf, Gaussian Mixture, ...
• Here we consider only discrete R.V.
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
41. Bayesian Networks Unit - Probabilistic Inference over Time p. 41
Sequence Diagram (1/4)
X P(X0)
S0 0.2
X0 X1 X2 Xt
S1 0.1
... ... E1 E2 Et
Si 0.3
P(X0) probability table ?
• Suppose the unobservable variable X
is a discrete R.V.
•X = S1, S2, S3, ..., Si, ...
• P(X0) is the probability of X=Si at t=0
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
42. Bayesian Networks Unit - Probabilistic Inference over Time p. 42
Sequence Diagram (2/4)
X0 X1 X2 Xt
E1 E2 Et
P(Xt|Xt-1) conditional probability table ?
Xt+1 S1 S2 ... Si
Xt
Transition S1 0.1 0.2 ... 0.05
probability S2 0.2 0.15 ... 0.18
... ... ... ... ...
Si 0.31 0.03 ... 0.22
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
43. Bayesian Networks Unit - Probabilistic Inference over Time p. 43
Sequence Diagram (3/4)
X0 X1 X2 Xt
E1 E2 Et
P(Et|Xt) conditional probability table ?
Et v1 v2 ... vj
Xt
Observation S1 0.1 0.2 ... 0.05
probability S2 0.2 0.15 ... 0.18
... ... ... ... ...
Si 0.31 0.03 ... 0.22
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
44. Bayesian Networks Unit - Probabilistic Inference over Time p. 44
Sequence Diagram (4/4)
X1 X2 X3 X4 X5 X6 X7
S3 S3 S1 S1 S3 S2 S3
v2 v4 v1 v1 v2 v3 v4
E1 E2 E3 E4 E5 E6 E7
t =1 2 3 4 5 6 7
S1 S1 S1 S1 S1 S1 S1
S2 S2 S2 S2 S2 S2 S2
S3 S3 S3 S3 S3 S3 S3
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
45. Bayesian Networks Unit - Probabilistic Inference over Time p. 45
Short Summary
X0 X1 X2 Xt
E1 E2 Et
• If we have the three PDFs/Tables
– P(X0), P(Xi|Xi-1), P(Ei|Xi)
• We can answer any query
– P(X1, X3 | E2, E4), P(X1, E5 | X2, X4), ...
• Do we need to ask many kinds of query?
• Or we have some frequently asked queries?
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
46. Bayesian Networks Unit - Probabilistic Inference over Time p. 46
2. Inference in Temporal Models
• Four common query tasks in
temporal inference/reasoning
– Filtering: P(Xt | e1:t)= P(Xt | E1:t=e1:t)
• Estimate correct current states
– Prediction: P(Xt+k | e1:t) for k > 0
• Predict possible next states
– Smoothing: P(Xk | e1:t) for 1 k < t
• Better estimate of past states
– Most likely explanation:
arg maxX1:t P(X1:t | e1:t)
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
47. Bayesian Networks Unit - Probabilistic Inference over Time p. 47
Subsections
• 2.1 Graphical models of the 4
inferences
• 2.2 Mathematical formula of the 4
inferences
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
48. Bayesian Networks Unit - Probabilistic Inference over Time p. 48
2.1 Graphical Models
of the 4 Inferences
• Use sequence diagram to illustrate
what are
– Filtering
– Prediction
– Smoothing
– Most likely explanation
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
49. Bayesian Networks Unit - Probabilistic Inference over Time p. 49
Graphical Models - Filtering
• P(Xt | e1:t) X0 X1 X2 Xt
E1 E2 Et
A filtering example for 2D
position of robot/WLAN
card
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
50. Bayesian Networks Unit - Probabilistic Inference over Time p. 50
Graphical Models - Prediction
• P(Xt+k | e1:t) for k > 0
For k=1
X0 X1 X2 Xt Xt+1
E1 E2 Et
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
51. Bayesian Networks Unit - Probabilistic Inference over Time p. 51
Graphical Models – Smoothing (1/3)
• P(Xk | e1:t) for 1 k < t
X0 X1 X2 Xk Xt
E1 E2 Ek Et
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
52. Bayesian Networks Unit - Probabilistic Inference over Time p. 52
Graphical Models – Smoothing (2/3)
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
53. Bayesian Networks Unit - Probabilistic Inference over Time p. 53
Graphical Models – Smoothing (3/3)
Smoothing v.s. Filtering
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
54. Bayesian Networks Unit - Probabilistic Inference over Time p. 54
Graphical Models
- Most Likely Explanation (1/2)
• arg maxX1:t P(X1:t | e1:t)
X0 X1 X2 Xt
E1 E2 Et
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
55. Bayesian Networks Unit - Probabilistic Inference over Time p. 55
Graphical Models
- Most Likely Explanation (2/2)
t =1 2 3 4 5 6 7
E1=v2 E2=v4 E3=v1 E4=v1 E5=v2 E6=v3 E7=v4
S1 S1 S1 S1 S1 S1 S1
S2 S2 S2 S2 S2 S2 S2
S3 S3 S3 S3 S3 S3 S3
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
56. Bayesian Networks Unit - Probabilistic Inference over Time p. 56
2.2 Mathematical Formula
of the 4 Inferences
• Derive mathematical formula of
– Prediction
– Filtering
– Smoothing
– Most likely explanation
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
57. Bayesian Networks Unit - Probabilistic Inference over Time p. 57
Prediction (1/3)
• P(Xt+1 | e1:t ): one-step prediction as
example
X0 X1 X2 Xt Xt+1
E1 E2 Et
P( X t 1 | e1:t )
X 0 , X t
P( X 0 ) P ( X t 1 | X t ) P( X i | X i 1 ) P(ei | X i )
i 1t
But more efficient formula can be derived
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
58. Bayesian Networks Unit - Probabilistic Inference over Time p. 58
Prediction (2/3)
• New formula for P(Xt+1 | e1:t )
– Xt+1 has no relationship to e1, e2, ..., et
– But they both have relationship to xt
– If X is a Boolean R.V., P(Xt+1|e1:t)
=<P(xt+1=true|e1:t), P(xt+1=false|e1:t)>
• P(Xt+1 | e1:t )
– = xt P(Xt+1 | xt , e1:t )P(xt | e1:t ) transition model
by
– = xt P(Xt+1 | xt )P(xt | e1:t )
CPT of transition model Filtering
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
59. Bayesian Networks Unit - Probabilistic Inference over Time p. 59
Prediction (3/3)
P(Xt+1 | e1:t) = xt P(Xt+1| xt )P(xt | e1:t )
t=1 2 3 ... t t+1
Et=v2 v4 v1 ... v3 e1:t
P(Xt+1=S2 | e1:t)
S1 S1 S1 S1 P(S1|S2) ) S1
= xt P(Xt+1=S2| xt=Si) … P(S1|e1:t
P(xt=Si | e1:t ) S2 P(S2|S2) S2
S2 S2 S2
= xt P(S2|Si)P(Si|e1:t) … P(S2|e1:t)
= P(S2|S1)P(S1|e1:t) S3 S3 S3 S3 P(S3|S2) S3
+ P(S2|S2)P(S2|e1:t) … P(S3|e1:t)
…
…
…
…
…
+ P(S2|S3)P(S3|e1:t) P(SN|S2)
+ ... SN SN SN … SN P(SN|e1:t) SN
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
60. Bayesian Networks Unit - Probabilistic Inference over Time p. 60
Filtering (1/3)
• P(Xt+1 | e1:t+1) X X1 X2 Xt+1
0
( or P(Xt | e1:t) )
E1 E2 Et+1
P( X t 1 | e1:t 1 )
X 0 , X t
P( X 0 ) P ( X i | X i 1 ) P (ei | X i )
i 1t
But more efficient formula can be derived
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
61. Bayesian Networks Unit - Probabilistic Inference over Time p. 61
Filtering (2/3)
• P(Xt+1 | e1:t+1)
• P(Xt+1 | e1:t+1)= P(Xt+1 | e1:t, et+1)
– = P(et+1 | Xt+1 , e1:t) P(Xt+1 | e1:t) by Bayes rule
– = P(et+1 | Xt+1 ) P(Xt+1 | e1:t ) by sensor model
CPT of sensor model Prediction by transition model
– = P(et+1 | Xt+1 ) xt P(Xt+1 | xt , e1:t )P(xt | e1:t )
– = P(et+1 | Xt+1 ) xt P(Xt+1 | xt )P(xt | e1:t )
• We derive a recursive algorithm
– P(Xt+1 | e1:t+1) can be calculated by P(Xt | e1:t)
– There is a function f that
P(Xt+1 | e1:t+1) = f(et+1, P(Xt | e1:t))
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
62. Bayesian Networks Unit - Probabilistic Inference over Time p. 62
Filtering (3/3)
P(Xt+1 | e1:t+1)= P(et+1 | Xt+1)P(Xt+1 | e1:t )
t=1 2 ... t t+1
Et=v2 v4 ... v3 v4
P(Xt+1=S2 | e1:t+1) P(S1|S2) S et+1
S1 S1 … S1 P(S |e ) 1
= P(et+1=v4|Xt+1=S2) 1 1:t
P(et+1|Xt+1)
P(Xt+1=S2 | e1:t ) S2 S2 … S2 P(S2|S2) S2 =P(v4|S2)
P(S2|e1:t) P(Xt+1|e1:t)
= P(v4 | S2) P(S3|S2) S =P(S2|e1:t)
P(S2 | e1:t ) S3 S3 … S 3 P(S
3|e1:t)
3
…
…
…
…
P(SN|S2)
SN SN … SN P(SN|e1:t) SN
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
63. Bayesian Networks Unit - Probabilistic Inference over Time p. 63
Forward Variable
• P(Xt+1 | e1:t ) = xt P(Xt+1 | xt )P(xt | e1:t )
• P(Xt+1 | e1:t+1) = P(et+1 | Xt+1) P(Xt+1 | e1:t )
• They are a kind of recursive function
• Interesting points
– Both the prediction & filtering of Xt+1 need
P(Xt|e1:t)
– We define the P(Xt|e1:t) as a forward
variable f1:t
– i.e., f1:t = P(Xt | e1:t ), f1:t(Si) = P(Xt=Si | e1:t )
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
64. Bayesian Networks Unit - Probabilistic Inference over Time p. 64
Forward Procedure
• P(Xt+1 | e1:t+1) = P(et+1 | Xt+1) P(Xt+1 | e1:t )
= P(et+1 | Xt+1) xt P(Xt+1 | xt )P(xt | e1:t )
• The filtering process is rewritten as
f1:t+1 = Forward(f1:t , et+1)
– A forward procedure (algorithm) :
Forward(f1:t , et+1)
= P(et+1 | Xt+1)xt P(Xt+1 | xt )P(xt | e1:t )
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
65. Bayesian Networks Unit - Probabilistic Inference over Time p. 65
Filtering Example (1/4)
• For the umbrella example
P(Rt|Rt-1)
P(Ut|Rt)
P(ut|rt)=0.9
P(ut|rt)=0.1
P(ut|rt)=0.2
P(ut|rt)=0.8
P(Ut|rt)=<P(ut|rt),P(ut|rt)> = <0.9,0.1>
P(Ut|rt)=<P(ut|rt),P(ut|rt)> = <0.2,0.8> P(ut|Rt)?
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
66. Bayesian Networks Unit - Probabilistic Inference over Time p. 66
Filtering Example (2/4)
• Assume the man believes that
P(R0) = <0.5,0.5> = < P(r0), P(r0) >
– The rain probability before the
observation sequence begins
• Now we has the observation sequence:
umbrella1=true, umbrella2=true
• We will use the P(R |U ) P(R2|U1,U2)
1 1
filtering process
Rain1 Rain2
to find rain
probability
Umbrella1 Umbrella2
=true =true
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
67. Bayesian Networks Unit - Probabilistic Inference over Time p. 67
Filtering Example (3/4)
Rain0 Rt-1 P(Rt)
Rain1 Rain2
Rt P(Ut)
P(R1|u1) t 0.7
t 0.9
f 0.3
= P(u1|R1)P(R1) f 0.2
Umbrella1 Umbrella2
P(R1) =true =true
= r0P(R1|r0)P(r0)
= <0.7,0.3>0.5 + <0.3,0.7>0.5 = <0.5,0.5>
P(R1|u1) = P(u1|R1)P(R1)
= <0.9,0.2><0.5,0,5> = <0.45,0.1>
<0.818, 0.182>
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
68. Bayesian Networks Unit - Probabilistic Inference over Time p. 68
Filtering Example (4/4)
Rain0 Rain1 Rain2
Rt-1 P(Rt) Rt P(Ut)
t 0.7 t 0.9
f 0.3 f 0.2
P(R2|u1,u2)
= P(u2|R2)P(R2|u1) Umbrella1 Umbrella2
=true =true
P(R2|u1)
= r1P(R2|r1)P(r1|u1)
= <0.7,0.3>0.818 + <0.3,0.7>0.182 = <0.627,0.373>
P(R2|u1,u2) = P(u2|R2)P(R2|u1)
= <0.9,0.2><0.627,0,373> = <0.565,0.075>
<0.883, 0.117>
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
69. Bayesian Networks Unit - Probabilistic Inference over Time p. 69
Smoothing (1/2)
• P(Xk | e1:t) for 1 k < t
– Divide e1:t into e1:k and ek+1:t
– P(Xk | e1:t) = P(Xk | e1:k , ek+1:t)
– = P(Xk | e1:k)P(ek+1:t | Xk , e1:k)
– = P(Xk | e1:k)P(ek+1:t | Xk )
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
70. Bayesian Networks Unit - Probabilistic Inference over Time p. 70
Smoothing (2/2)
P(Xk=S2 | e1:t)
t=1 k-1 k k+1 t
Et= v2 ... v4 v3 v1 ... v4
S1 … S1 P(x |x ) S1 S1 … S1
1 2
S2 … S2 P(x2|x2) S2 S2 … S2
P(x3|x2)
S3 … S3 S3 S3 … S3
…
…
…
…
…
P(xN|x2)
SN … SN SN SN … SN
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
71. Bayesian Networks Unit - Probabilistic Inference over Time p. 71
Backward Variable
• P(ek+1:t | Xk)
– = xk+1P(ek+1:t | Xk, xk+1)P(xk+1 | Xk)
– = xk+1P(ek+1:t | xk+1)P(xk+1 | Xk)
– = xk+1P(ek+1 , ek+2:t | xk+1)P(xk+1 | Xk)
– = xk+1P(ek+1 | xk+1)P(ek+2:t | xk+1)P(xk+1 | Xk)
• This is also a recursive formula
• We define a backward variable bk+1:t
– bk+1:t = P(ek+1:t | Xk)
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
72. Bayesian Networks Unit - Probabilistic Inference over Time p. 72
Backward Procedure (1/2)
• P(ek+1:t | Xk)
= xk+1P(ek+1|xk+1)P(ek+2:t|xk+1)P(xk+1|Xk)
• The formula is rewritten as
bk+1:t = Backward(bk+2:t , ek+1)
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
73. Bayesian Networks Unit - Probabilistic Inference over Time p. 73
Backward Procedure (2/2)
P(ek+1:t |xk) = xk+1P(ek+1|xk+1)P(ek+2:t|xk+1)P(xk+1|Xk)
t=1 k k+1 t
P(ek+1:t |xk=S2) Et= v2 ... v4 v1 ... v4
= xk+1P(v1|xk+1) S1 … S1 S1 … S1
P(ek+2:t|xk+1)P(xk+1|S2)
= P(v |S )P(e |S )P(S |S )
S2 … S2 S2 … S2
1 1 k+2:t 1 1 2
+ P(v1|S2)P(ek+2:t|S2)P(S2|S2)
+ ...
S3 … S3 S3 … S3
…
…
…
…
+ P(v1|SN)P(ek+2:t|SN)P(SN|S2)
SN … SN SN … SN
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
74. Bayesian Networks Unit - Probabilistic Inference over Time p. 74
The Smoothing Formula
• P(Xk | e1:t) = P(Xk | e1:k , ek+1:t)
– = P(Xk | e1:k)P(ek+1:t | Xk , e1:k)
– = P(Xk | e1:k)P(ek+1:t | Xk )
– = f1:kbk+1:t
• Time complexity
– Both the forward and backward
recursions take a constant time per
step
– Complexity of smoothing P(Xk | e1:t)
with e1:t is O(t)
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
75. Bayesian Networks Unit - Probabilistic Inference over Time p. 75
Smoothing Example (1/3)
• For the umbrella example
• P(R1 | u1, u2)
– Computing the smoothed estimate for the
probability of rain at t=1,
– Given the umbrella observations on days 1 & 2
Rain0 Rain1 Rain2
Rt-1 P(Rt) Rt P(Ut)
t 0.7 t 0.9
f 0.3 f 0.2
Umbrella1 Umbrella2
=true =true
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
76. Bayesian Networks Unit - Probabilistic Inference over Time p. 76
Smoothing Example (2/3)
• P(R1 | u1, u2) = P(R1|u1)P(u2|R1)
– P(R1|u1) = <0.818, 0.182>
– P(u2|R1) = r2P(u2|r2)P(|r2)P(r2|R1)
= (0.91<0.7,0.3>) + (0.21<0.3,0.7>)
= <0.69, 0.41>
• P(R1 | u1, u2) = <0.818,0.182><0.69,0.41>
<0.883, 0.117>
• Note: P(R1|u1) = <0.818, 0.182>
• With more one observation u2, the
probability of r1 increases smoothing
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
77. Bayesian Networks Unit - Probabilistic Inference over Time p. 77
Smoothing Example (3/3)
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
78. Bayesian Networks Unit - Probabilistic Inference over Time p. 78
Most Likely Explanation (1/2)
• Smoothing P(Xk | e1:t) considers only
one past state at time step k
• Most likely explanation,
arg maxX1:t P(X1:t | e1:t)
– Considers all past states, and
– Choose the best state sequence
X0 X1 X2 Xt
E1 E2 Et
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
79. Bayesian Networks Unit - Probabilistic Inference over Time p. 79
Most Likely Explanation (2/2)
• We will discuss 3 algorithms
– Algorithm 1:
• Very simple, directly using smoothing
• Time complexity O(t2)
– Algorithm 2(forward-backward algo.):
• Improved usage of smoothing
• Time complexity O(t)
• But the result may not be the best state
sequence
– Algorithm 3(Viterbi algorithm):
• Time complexity O(t)
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
80. Bayesian Networks Unit - Probabilistic Inference over Time p. 80
Algorithm 1
• The most simple idea for this
problem
– Call smoothing t times, smoothing one
state each time
– For (i=0; i<t; i++) P(Xi | e1:t)
• Drawback
– Time complexity of O(t2) : too slow
• Improvement
– Apply dynamic programming to
reduce the complexity to O(t)
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
81. Bayesian Networks Unit - Probabilistic Inference over Time p. 81
Algorithm 2 (1/2)
• Forward-backward algorithm
– First, record the results of the forward
filtering over the whole sequence from 1
to t
– Then, run the backward recursion from t
down to 1, and
• Compute the smoothed estimate at each time
step k, from the bk+1:t and the stored f1:k
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
82. Bayesian Networks Unit - Probabilistic Inference over Time p. 82
Algorithm 2 (2/2)
fv[i]= f1:t = P(Xt | e1:t )
forward procedure: f1:t+1 = Forward(f1:t , et+1)
Smoothing: P(Xk | e1:t) = f1:kbk+1:t
backward procedure: bk+1:t = Backward(bk+2:t , ek+1)
Fu Jen University in previous slidesElectrical Engineering
Department of Wang, Yuan-Kai Copyright
83. Bayesian Networks Unit - Probabilistic Inference over Time p. 83
However (1/2)
• For the umbrella example, suppose there is
an observation sequence e1:t=[true, true,
false, true, true] for umbrella's appearance
• What is the weather sequence most likely to
explain this?
– Does the absence of the umbrella on day 3
mean that
• Day 3 wasn't raining, or
• The director forget to bring it?
• If day 3 wasn't raining, day 4 may not be raining
either, but the director brought the umbrella just
in case
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
84. Bayesian Networks Unit - Probabilistic Inference over Time p. 84
However (2/2)
• The forward-backward algorithm
uses smoothing for each single time
step
• But to find the most likely sequence,
we must consider joint probabilities
over all time steps
• To consider joint probabilities of a
sequence, we need to consider path
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
85. Bayesian Networks Unit - Probabilistic Inference over Time p. 85
Path
• A path is a possible sequence
– There are 25 paths
– Each path (sequence) has a probability
– Only one path has the maximum
probability
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
86. Bayesian Networks Unit - Probabilistic Inference over Time p. 86
Probability of Path
t
P( X 1:t | e1:t ) P( X i 1 | X i ) P(ei | X i )
i 1
• arg maxX1:t P(X1:t | e1:t)
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
87. Bayesian Networks Unit - Probabilistic Inference over Time p. 87
Recursive View
• An important idea for finding
arg maxX1:t P(X1:t | e1:t)
– A path in maxX1:t-1 P(X1:t-1 | e1:t-1) must
be the path in maxX1:t P(X1:t | e1:t)
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
88. Bayesian Networks Unit - Probabilistic Inference over Time p. 88
The Viterbi Example
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
89. Bayesian Networks Unit - Probabilistic Inference over Time p. 89
Algorithm 3
• Viterbi algorithm
max P( x1 , , xt , X t 1 | e1:t 1 )
x1 xt
P (et 1 | X t 1 ) max P( X t 1 | xt ) max P ( x1 , , xt 1 , xt | e1:t )
xt x1 xt 1
• It is similar to the filtering algorithm
P(Xt+1 | e1:t+1)
= P(et+1 | Xt+1) xt P(Xt+1 | xt )P(xt | e1:t )
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
90. Bayesian Networks Unit - Probabilistic Inference over Time p. 90
3. Various Models
• Hidden Markov Models
• Kalman Filter
• Particle Filter
• Dynamic Bayesian Networks
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
91. Bayesian Networks Unit - Probabilistic Inference over Time p. 91
Hidden Markov Model (1/2)
Hidden states
eg. Real location X1 X2 X3
Observations
eg. Detected Y1 Y2 Y3
location
n
P( x1 , x2 ,..., xn ) P( xi | pa( xi ))
i 1
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
92. Bayesian Networks Unit - Probabilistic Inference over Time p. 92
Hidden Markov Model (2/2)
X1 A
X2 X3
Parameter tyeing
B Y1 Y3
Y2
Transition matrix
Observation matrix
Initial state distribution
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
93. Bayesian Networks Unit - Probabilistic Inference over Time p. 93
Kalman Filtering
X1 X2 X3
Y1 Y2 Y3
• The same graphical structure with HMM
• But
•In HMM, Xi and Yi are discrete (CPT)
•In Kalman filter, Xi and Yi are continuous
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
94. Bayesian Networks Unit - Probabilistic Inference over Time p. 94
Particle Filtering
• TBU
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
95. Bayesian Networks Unit - Probabilistic Inference over Time p. 95
Dynamic Bayesian Network (DBN)
• TBU
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright
96. Bayesian Networks Unit - Probabilistic Inference over Time p. 96
4. References
• Chapter 15, Sections 15.1-15.2, Artificial
Intelligence - a modern approach, 2nd, by S.
Russel & P. Norvig, Prentice Hall, 2003.
Fu Jen University Department of Electrical Engineering Wang, Yuan-Kai Copyright