PROBABILISTIC REASONING OVER TIME

Transcription

1 PROBABILISTIC REASONING OVER TIME In which we try to interpret the present, understand the past, and perhaps predict the future, even when very little is crystal clear.

2 Outline Time and uncertainty Inference: filtering, prediction, smoothing Hidden Markov models Kalman filters (a brief mention) Dynamic Bayesian networks Particle filtering Chapter 15, Sections 1 5 2

3 Time and uncertainty The world changes; we need to track and predict it Diabetes management vs vehicle diagnosis Basic idea: copy state and evidence variables for each time step X t =setofunobservablestatevariablesattimet e.g., BloodSugar t, StomachContents t,etc. E t =setofobservableevidencevariablesattimet e.g., MeasuredBloodSugar t, PulseRate t, FoodEaten t This assumes discrete time; step size depends on problem Notation: X a:b = X a, X a+1,...,x b 1, X b Chapter 15, Sections 1 5 3

4

5 Markov processes (Markov chains) Construct a Bayes net from these variables: parents? Markov assumption: X t depends on bounded subset of X 0:t 1 First-order Markov process: P(X t X 0:t 1 )=P(X t X t 1 ) Second-order Markov process: P(X t X 0:t 1 )=P(X t X t 2, X t 1 ) First order X t 2 X t 1 X t X t +1 X t +2 Second order X t 2 X t 1 X t X t +1 X t +2 Sensor Markov assumption: P(E t X 0:t, E 0:t 1 )=P(E t X t ) Stationary process: transition model P(X t X t 1 ) and sensor model P(E t X t ) fixed for all t Chapter 15, Sections 1 5 4

6 Example Rain t 1 R t 1 t f P(R t ) Rain t Rain t +1 Rt t f P(U t ) Umbrella t 1 Umbrella t Umbrella t +1 First-order Markov assumption not exactly true in real world! Possible fixes: 1. Increase order of Markov process 2. Augment state, e.g.,addt emp t, Pressure t Example: robot motion. Augment position and velocity with Battery t Chapter 15, Sections 1 5 5

7

8

9 Inference tasks Filtering: P(X t e 1:t ) belief state input to the decision process of a rational agent Prediction: P(X t+k e 1:t ) for k>0 evaluation of possible action sequences; like filtering without the evidence Smoothing: P(X k e 1:t ) for 0 k<t better estimate of past states, essential for learning Most likely explanation: arg max x1:t P (x 1:t e 1:t ) speech recognition, decoding with a noisy channel Chapter 15, Sections 1 5 6

10 Filtering Aim: devise a recursive state estimation algorithm: P(X t+1 e 1:t+1 )=f(e t+1, P(X t e 1:t )) P(X t+1 e 1:t+1 )=P(X t+1 e 1:t, e t+1 ) = αp(e t+1 X t+1, e 1:t )P(X t+1 e 1:t ) = αp(e t+1 X t+1 )P(X t+1 e 1:t ) I.e.,prediction + estimation. PredictionbysummingoutX t : P(X t+1 e 1:t+1 )=αp(e t+1 X t+1 )Σ xt P(X t+1 x t, e 1:t )P (x t e 1:t ) = αp(e t+1 X t+1 )Σ xt P(X t+1 x t )P (x t e 1:t ) f 1:t+1 = Forward(f 1:t, e t+1 ) where f 1:t = P(X t e 1:t ) Time and space constant (independent of t) Chapter 15, Sections 1 5 7

11 Filtering example True False Rain 0 Rain 1 Rain 2 Umbrella 1 Umbrella 2 Chapter 15, Sections 1 5 8

12 Filtering example True False Rain 0 Rain 1 Rain 2 Umbrella 1 Umbrella 2 Chapter 15, Sections 1 5 8

13 Filtering example True False Rain 0 Rain 1 Rain 2 Umbrella 1 Umbrella 2 Chapter 15, Sections 1 5 8

14

15 Smoothing example True False forward smoothed backward Rain 0 Rain 1 Rain 2 Umbrella 1 Umbrella 2 Forward backward algorithm: cache forward messages along the way Time linear in t (polytree inference), space O(t f ) Chapter 15, Sections

16 Smoothing example True False forward smoothed backward Rain 0 Rain 1 Rain 2 Umbrella 1 Umbrella 2 Forward backward algorithm: cache forward messages along the way Time linear in t (polytree inference), space O(t f ) Chapter 15, Sections

17 Most likely explanation Most likely sequence sequence of most likely states!!!! Most likely path to each x t+1 =mostlikelypathtosome x t plus one more step max P(x x 1...x 1,...,x t, X t+1 e 1:t+1 ) t = P(e t+1 X t+1 )max P(Xt+1 x x t ) max P (x t x 1...x 1,...,x t 1, x t e 1:t ) t 1 Identical to filtering, except f 1:t replaced by m 1:t = max P(x x 1...x 1,...,x t 1, X t e 1:t ), t 1 I.e., m 1:t (i) gives the probability of the most likely path to state i. Update has sum replaced by max, giving the Viterbi algorithm: m 1:t+1 = P(e t+1 X t+1 )max x t (P(X t+1 x t )m 1:t ) Chapter 15, Sections

18 Viterbi example Rain 1 Rain 2 Rain 3 Rain 4 Rain 5 state space paths true false true false true false true false true false umbrella true true false true true most likely paths m 1:1 m 1:2 m 1:3 m 1:4 m 1:5 Chapter 15, Sections

19

20 Hidden Markov models X t is a single, discrete variable (usually E t is too) Domain of X t is {1,...,S} Transition matrix T ij = P (X t = j X t 1 = i), e.g., Sensor matrix O t for each time step, diagonal elements P (e t X t = i) e.g., with U 1 = true, O 1 = Forward and backward messages as column vectors: f 1:t+1 = αo t+1 T f 1:t b k+1:t = TO k+1 b k+2:t Forward-backward algorithm needs time O(S 2 t) and space O(St) Chapter 15, Sections

21 Hidden Markov models X t is a single, discrete variable (usually E t is too) Domain of X t is {1,...,S} Transition matrix T ij = P (X t = j X t 1 = i), e.g., Sensor matrix O t for each time step, diagonal elements P (e t X t = i) e.g., with U 1 = true, O 1 = Forward and backward messages as column vectors: f 1:t+1 = αo t+1 T f 1:t b k+1:t = TO k+1 b k+2:t Forward-backward algorithm needs time O(S 2 t) and space O(St) Chapter 15, Sections

22 Hidden Markov models X t is a single, discrete variable (usually E t is too) Domain of X t is {1,...,S} Transition matrix T ij = P (X t = j X t 1 = i), e.g., Sensor matrix O t for each time step, diagonal elements P (e t X t = i) e.g., with U 1 = true, O 1 = Forward and backward messages as column vectors: f 1:t+1 = αo t+1 T f 1:t b k+1:t = TO k+1 b k+2:t Forward-backward algorithm needs time O(S 2 t) and space O(St) Chapter 15, Sections

23

24 Country dance algorithm Can avoid storing all forward messages in smoothing by running forward algorithm backwards: f 1:t+1 = αo t+1 T f 1:t O 1 t+1f 1:t+1 = αt f 1:t α (T ) 1 O 1 t+1f 1:t+1 = f 1:t Algorithm: forward pass computes f t,backwardpassdoesf i, b i Chapter 15, Sections

25 Country dance algorithm Can avoid storing all forward messages in smoothing by running forward algorithm backwards: f 1:t+1 = αo t+1 T f 1:t O 1 t+1f 1:t+1 = αt f 1:t α (T ) 1 O 1 t+1f 1:t+1 = f 1:t Algorithm: forward pass computes f t,backwardpassdoesf i, b i Chapter 15, Sections

26 Country dance algorithm Can avoid storing all forward messages in smoothing by running forward algorithm backwards: f 1:t+1 = αo t+1 T f 1:t O 1 t+1f 1:t+1 = αt f 1:t α (T ) 1 O 1 t+1f 1:t+1 = f 1:t Algorithm: forward pass computes f t,backwardpassdoesf i, b i Chapter 15, Sections

27 Country dance algorithm Can avoid storing all forward messages in smoothing by running forward algorithm backwards: f 1:t+1 = αo t+1 T f 1:t O 1 t+1f 1:t+1 = αt f 1:t α (T ) 1 O 1 t+1f 1:t+1 = f 1:t Algorithm: forward pass computes f t,backwardpassdoesf i, b i Chapter 15, Sections

28 Country dance algorithm Can avoid storing all forward messages in smoothing by running forward algorithm backwards: f 1:t+1 = αo t+1 T f 1:t O 1 t+1f 1:t+1 = αt f 1:t α (T ) 1 O 1 t+1f 1:t+1 = f 1:t Algorithm: forward pass computes f t,backwardpassdoesf i, b i Chapter 15, Sections

29 Country dance algorithm Can avoid storing all forward messages in smoothing by running forward algorithm backwards: f 1:t+1 = αo t+1 T f 1:t O 1 t+1f 1:t+1 = αt f 1:t α (T ) 1 O 1 t+1f 1:t+1 = f 1:t Algorithm: forward pass computes f t,backwardpassdoesf i, b i Chapter 15, Sections

30 Country dance algorithm Can avoid storing all forward messages in smoothing by running forward algorithm backwards: f 1:t+1 = αo t+1 T f 1:t O 1 t+1f 1:t+1 = αt f 1:t α (T ) 1 O 1 t+1f 1:t+1 = f 1:t Algorithm: forward pass computes f t,backwardpassdoesf i, b i Chapter 15, Sections

31 Country dance algorithm Can avoid storing all forward messages in smoothing by running forward algorithm backwards: f 1:t+1 = αo t+1 T f 1:t O 1 t+1f 1:t+1 = αt f 1:t α (T ) 1 O 1 t+1f 1:t+1 = f 1:t Algorithm: forward pass computes f t,backwardpassdoesf i, b i Chapter 15, Sections

32 Country dance algorithm Can avoid storing all forward messages in smoothing by running forward algorithm backwards: f 1:t+1 = αo t+1 T f 1:t O 1 t+1f 1:t+1 = αt f 1:t α (T ) 1 O 1 t+1f 1:t+1 = f 1:t Algorithm: forward pass computes f t,backwardpassdoesf i, b i Chapter 15, Sections

33 Country dance algorithm Can avoid storing all forward messages in smoothing by running forward algorithm backwards: f 1:t+1 = αo t+1 T f 1:t O 1 t+1f 1:t+1 = αt f 1:t α (T ) 1 O 1 t+1f 1:t+1 = f 1:t Algorithm: forward pass computes f t,backwardpassdoesf i, b i Chapter 15, Sections

34

35

36

37

38 Kalman filters Modelling systems described by a set of continuous variables, e.g., tracking a bird flying X t = X, Y, Z, Ẋ,Ẏ,Ż. Airplanes, robots, ecosystems, economies, chemical plants, planets,... Xt Xt+1 Xt Xt+1 Zt Zt+1 Gaussian prior, linear Gaussian transition model and sensor model Chapter 15, Sections

39 Updating Gaussian distributions Prediction step: if P(X t e 1:t ) is Gaussian, then prediction P(X t+1 e 1:t )= x t P(X t+1 x t )P (x t e 1:t ) dx t is Gaussian. If P(X t+1 e 1:t ) is Gaussian, then the updated distribution P(X t+1 e 1:t+1 )=αp(e t+1 X t+1 )P(X t+1 e 1:t ) is Gaussian Hence P(X t e 1:t ) is multivariate Gaussian N(µ t, Σ t ) for all t General (nonlinear, non-gaussian) process: description of posterior grows unboundedly as t Chapter 15, Sections

40 Simple 1-D example Gaussian random walk on X axis, s.d. σ x,sensors.d.σ z µ t+1 = (σ2 t + σ 2 x)z t+1 + σ 2 zµ t σ 2 t + σ 2 x + σ 2 z σ 2 t+1 = (σ2 t + σ 2 x)σ 2 z σ 2 t + σ 2 x + σ 2 z P(X) P(x1 z1=2.5) 0.25 P(x0) P(x1) *z X position Chapter 15, Sections

41 Transition and sensor models: General Kalman update P (x t+1 x t )=N(Fx t, Σ x )(x t+1 ) P (z t x t )=N(Hx t, Σ z )(z t ) F is the matrix for the transition; Σ x the transition noise covariance H is the matrix for the sensors; Σ z the sensor noise covariance Filter computes the following update: µ t+1 = Fµ t + K t+1 (z t+1 HFµ t ) Σ t+1 =(I K t+1 )(FΣ t F + Σ x ) where K t+1 =(FΣ t F + Σ x )H (H(FΣ t F + Σ x )H + Σ z ) 1 is the Kalman gain matrix Σ t and K t are independent of observation sequence, so compute offline Chapter 15, Sections

42 2-D tracking example: filtering 12 2D filtering 11 true observed filtered 10 Y X Chapter 15, Sections

43 2-D tracking example: smoothing 12 2D smoothing 11 true observed smoothed 10 Y X Chapter 15, Sections

44 Where it breaks Cannot be applied if the transition model is nonlinear Extended Kalman Filter models transition as locally linear around x t = µ t Fails if systems is locally unsmooth Chapter 15, Sections

45 Dynamic Bayesian networks X t, E t contain arbitrarily many variables in a replicated Bayes net BMeter 1 P(R ) R 0 t f P(R 1) Battery 0 Battery 1 Rain 0 Rain 1 R 1 t f P(U 1) X 0 X 1 Umbrella 1 txx 0 X 1 Z 1 Chapter 15, Sections

46 DBNs vs. HMMs Every HMM is a single-variable DBN; every discrete DBN is an HMM X t X t+1 Y t Yt+1 Z t Z t+1 Sparse dependencies exponentially fewer parameters; e.g., 20 state variables, three parents each DBN has =160 parameters, HMM has Chapter 15, Sections

47 DBNs vs Kalman filters Every Kalman filter model is a DBN, but few DBNs are KFs; real world requires non-gaussian posteriors E.g., where are bin Laden and my keys? What s the battery charge? BMBroken0 BMBroken1 BMeter 1 Battery 0 Battery 1 5 E(Battery ) 4 E(Battery ) X 0 txx 0 X 1 X 1 E(Battery) P(BMBroken ) 0 P(BMBroken ) Z Time step Chapter 15, Sections

48 Exact inference in DBNs Naive method: unroll the network and run any exact algorithm P(R 0) 0.7 Rain 0 R 0 t f P(R 1) Rain 1 P(R 0) 0.7 Rain 0 R 0 t f P(R 1) Rain 1 R 0 t f P(R 1) Rain 2 R 0 t f P(R 1) Rain 3 R 0 t f P(R 1) Rain 4 R 0 t f P(R 1) Rain 5 R 0 t f P(R 1) Rain 6 R 0 t f P(R 1) Rain 7 R 1 t f P(U 1) R 1 t f P(U 1) R 1 t f P(U 1) R 1 t f P(U 1) R 1 t f P(U 1) R 1 t f P(U 1) R 1 t f P(U 1) R 1 t f P(U 1) Umbrella 1 Umbrella 1 Umbrella 2 Umbrella 3 Umbrella 4 Umbrella 5 Umbrella 6 Umbrella 7 Problem: inference cost for each update grows with t Rollup filtering: add slice t +1, sum out slice t using variable elimination Largest factor is O(d n+1 ), update cost O(d n+2 ) (cf. HMM update cost O(d 2n )) Chapter 15, Sections

49 Likelihood weighting for DBNs Set of weighted samples approximates the belief state Rain 0 Rain 1 Rain 2 Rain 3 Rain 4 Rain 5 Umbrella 1 Umbrella 2 Umbrella 3 Umbrella 4 Umbrella 5 LW samples pay no attention to the evidence! fraction agreeing falls exponentially with t number of samples required grows exponentially with t LW(10) LW(100) LW(1000) LW(10000) RMS error Time step Chapter 15, Sections

50 Particle filtering Basic idea: ensure that the population of samples ( particles ) tracks the high-likelihood regions of the state-space Replicate particles proportional to likelihood for e t true Rain t Rain t +1 Rain t +1 Rain t +1 false (a) Propagate (b) Weight (c) Resample Widely used for tracking nonlinear systems, esp. in vision Also used for simultaneous localization and mapping in mobile robots dimensional state space Chapter 15, Sections

51 Particle filtering contd. Assume consistent at time t: N(x t e 1:t )/N = P (x t e 1:t ) Propagate forward: populations of x t+1 are N(x t+1 e 1:t )=Σ xt P (x t+1 x t )N(x t e 1:t ) Weight samples by their likelihood for e t+1 : W (x t+1 e 1:t+1 )=P (e t+1 x t+1 )N(x t+1 e 1:t ) Resample to obtain populations proportional to W : N(x t+1 e 1:t+1 )/N = αw (x t+1 e 1:t+1 )=αp (e t+1 x t+1 )N(x t+1 e 1:t ) = αp (e t+1 x t+1 )Σ xt P (x t+1 x t )N(x t e 1:t ) = α P (e t+1 x t+1 )Σ xt P (x t+1 x t )P (x t e 1:t ) = P (x t+1 e 1:t+1 ) Chapter 15, Sections

52 Particle filtering performance Approximation error of particle filtering remains bounded over time, at least empirically theoretical analysis is difficult Avg absolute error LW(25) LW(100) LW(1000) LW(10000) ER/SOF(25) Time step Chapter 15, Sections

53 Summary Temporal models use state and sensor variables replicated over time Markov assumptions and stationarity assumption, so we need transition modelp(x t X t 1 ) sensor model P(E t X t ) Tasks are filtering, prediction, smoothing, most likely sequence; all done recursively with constant cost per time step Hidden Markov models have a single discrete state variable; used for speech recognition Kalman filters allow n state variables, linear Gaussian, O(n 3 ) update Dynamic Bayes nets subsume HMMs, Kalman filters; exact update intractable Particle filtering is a good approximate filtering algorithm for DBNs Chapter 15, Sections

54

55

56

57

58

59

60

61

62