On the stability and the applications of interacting particle systems

Size: px
Start display at page:

Download "On the stability and the applications of interacting particle systems"

Transcription

1 1/42 On the stability and the applications of interacting particle systems P. Del Moral INRIA Bordeaux Sud Ouest PDE/probability interactions - CIRM, April 2017 Synthesis joint works with A.N. Bishop, J. Garnier, A. Guionnet, A. Kurtzmann, J. Jacod M. Ledoux, L. Miclo, J. Tugaut, L. Wu

2 2/42 Nonlinear Markov chains Mean field particle methods Some stochastic perturbation tools Feynman-Kac models ( filtering, rare events,... ) A brief review on genetic type models Non commutative models Lyapunov weighted dynamics model Particle Gibbs-Glauber dynamics Many body Feynman-Kac (particle) models Duality formulae + Taylor expansions Branching proc. and Interacting MCMC Continuous time models Feynman-Kac models - Diffusion Monte Carlo Gradient type diffusions Ensemble Kalman-Bucy filter Stability + Unif. prop. chaos

3 3/42 Nonlinear Markov chains Mean field particle methods Some stochastic perturbation tools Feynman-Kac models ( filtering, rare events,... ) Particle Gibbs-Glauber dynamics Continuous time models

4 4/42 Nonlinear distribution flows Probability measures η n on some state spaces S n η n = Φ n (η n 1 )

5 4/42 Nonlinear distribution flows Probability measures η n on some state spaces S n η n = Φ n (η n 1 ) Ex. Feynman-Kac measures, conditional distributions, nonlinear MCMC, discrete time McKean-Vlasov, discrete time approximations of any nonlinear integro-diff eq. of proba (jumps, collisions, diffusions),...

6 4/42 Nonlinear distribution flows Probability measures η n on some state spaces S n η n = Φ n (η n 1 ) Ex. Feynman-Kac measures, conditional distributions, nonlinear MCMC, discrete time McKean-Vlasov, discrete time approximations of any nonlinear integro-diff eq. of proba (jumps, collisions, diffusions),... Continuous time Discrete time on time mesh Drift + Diffusion : Euler type scheme Jumps + Exponential rates : Geometric/Bernoulli type schemes η n = Φ n (η n 1 ) dt 0 nonlinear integro-diff equation

7 5/42 Nonlinear Markov models η n = Φ n (η n 1 ) Always K n,η Markov transitions s.t. Φ n (η n 1 ) = η n 1 K n,ηn 1 = Law ( ) X n = ηn i.e. : P(X n dx n X n 1 ) = K n,ηn 1 (X n 1, dx n )

8 5/42 Nonlinear Markov models η n = Φ n (η n 1 ) Always K n,η Markov transitions s.t. Φ n (η n 1 ) = η n 1 K n,ηn 1 = Law ( ) X n = ηn i.e. : P(X n dx n X n 1 ) = K n,ηn 1 (X n 1, dx n ) Non unique Markov interpretations,...

9 6/42 McKean measures (historical process) P( ( X 0,..., X n ) d(x0,..., x n )) = η 0 (dx 0 ) K 1,η0 (x 0, dx 1 )... K n,ηn 1 (x n 1, dx n )

10 6/42 McKean measures (historical process) P( ( ) X 0,..., X n d(x0,..., x n )) = η 0 (dx 0 ) K 1,η0 (x 0, dx 1 )... K n,ηn 1 (x n 1, dx n ) defined sequentially with η n = Φ n (η n 1 ) = n-th marginal = Law(X n )

11 6/42 McKean measures (historical process) P( ( ) X 0,..., X n d(x0,..., x n )) = η 0 (dx 0 ) K 1,η0 (x 0, dx 1 )... K n,ηn 1 (x n 1, dx n ) defined sequentially with η n = Φ n (η n 1 ) = n-th marginal = Law(X n ) Nb: the historical process is also a nonlinear Markov chain X n := ( X 0,..., X n )

12 6/42 McKean measures (historical process) P( ( ) X 0,..., X n d(x0,..., x n )) = η 0 (dx 0 ) K 1,η0 (x 0, dx 1 )... K n,ηn 1 (x n 1, dx n ) defined sequentially with η n = Φ n (η n 1 ) = n-th marginal = Law(X n ) Nb: the historical process is also a nonlinear Markov chain X n := ( X 0,..., X n ) Law(X n ) = η n = Φ n (η n 1 )

13 /42 McKean measures (historical process) P( ( ) X 0,..., X n d(x0,..., x n )) = η 0 (dx 0 ) K 1,η0 (x 0, dx 1 )... K n,ηn 1 (x n 1, dx n ) defined sequentially with η n = Φ n (η n 1 ) = n-th marginal = Law(X n ) Nb: the historical process is also a nonlinear Markov chain X n := ( X 0,..., X n ) Law(X n ) = η n = Φ n (η n 1 ) = η n 1 K n,ηn 1,...

14 7/42 Nonlinear Markov chains Mean field particle methods Some stochastic perturbation tools Feynman-Kac models ( filtering, rare events,... ) Particle Gibbs-Glauber dynamics Continuous time models

15 8/42 Mean field particle interpretation Markov chain ξ n = (ξ 1 n, ξ 2 n,..., ξ N n ) S N n ξ i n 1 ξ i n K n,η N n 1 (ξ i n 1, dx n ) with 1 N 1 i N δ ξ i n := η N n N η n Same models for the McKean measures ξ i n = (ξi 0,..., ξ i n) mean field particle of X n := ( X 0,..., X n )

16 8/42 Mean field particle interpretation Markov chain ξ n = (ξ 1 n, ξ 2 n,..., ξ N n ) S N n ξ i n 1 ξ i n K n,η N n 1 (ξ i n 1, dx n ) with 1 N 1 i N δ ξ i n := η N n N η n Same models for the McKean measures ξ i n = (ξi 0,..., ξ i n) mean field particle of X n := ( X 0,..., X n ) Interpretations: Stochastic adaptive grid, stochastic linearization, microscopic particle model, Sequential Monte Carlo method,...

17 9/42 Nonlinear Markov chains Mean field particle methods Some stochastic perturbation tools Feynman-Kac models ( filtering, rare events,... ) Particle Gibbs-Glauber dynamics Continuous time models

18 10/42 General fluctuation theorem local fluctuation random fields η N n Φ n ( η N n 1 ) = 1 N V N n

19 10/42 General fluctuation theorem local fluctuation random fields η N n Φ n ( η N n 1 ) = 1 N V N n f /Multivariate/Functional/Donsker theorems (dp+guionnet [99],Miclo[00],Ledoux [00],Jacod[02],...,Wu[11,13]) ( V N n ) n 0 N (V n ) n 0 with independent centered Gaussian fields V n s.t. E ( [V n (f )] 2) = η n 1 [ Kn,ηn 1 [f K n,ηn 1 (f )] 2]

20 11/42 Intuitive picture nonlinear sg : η n = Φ n(η n 1) = Φ p,n(η p) = η n Local fluctuations transport formulation : η 0 η 1 = Φ 1(η 0) η 2 = Φ 0,2(η 0) Φ 0,n(η 0) η0 N Φ 1(η0 N ) Φ 0,2(η0 N ) Φ 0,n(η0 N ) η1 N Φ 2(η1 N ) Φ 1,n(η1 N ) η2 N Φ 2,n(η2 N ). ηn 1 N Φ n(ηn 1) N ηn N

21 11/42 Intuitive picture nonlinear sg : η n = Φ n(η n 1) = Φ p,n(η p) = η n Local fluctuations transport formulation : η 0 η 1 = Φ 1(η 0) η 2 = Φ 0,2(η 0) Φ 0,n(η 0) η0 N Φ 1(η0 N ) Φ 0,2(η0 N ) Φ 0,n(η0 N ) η1 N Φ 2(η1 N ) Φ 1,n(η1 N ) η2 N Φ 2,n(η2 N ). ηn 1 N Φ n(ηn 1) N ηn N Local fluctuation errors η N n = Φ n(η N n 1) + 1 N V N n

22 12/42 Key decomposition formulae For any time-window ηn N η n = ηn N Φ n p,n(ηn p) N + Φ n p,n(ηn p) N Φ n p,n(η n p) }{{}}{{} e c 2 p then c = c 1 + c 2 e c 1 p / N p = (2c) 1 log N = e c 1 p / N = e c 2 p = More refined analysis ( 1 N ) c2 /(c 1 +c 2 ) η N n η n = n q=0 [Φ q,n(η N q ) Φ q,n(φ q(η N q 1))]

23 12/42 Key decomposition formulae For any time-window ηn N η n = ηn N Φ n p,n(ηn p) N + Φ n p,n(ηn p) N Φ n p,n(η n p) }{{}}{{} e c 2 p then c = c 1 + c 2 e c 1 p / N p = (2c) 1 log N = e c 1 p / N = e c 2 p = More refined analysis ( 1 N ) c2 /(c 1 +c 2 ) η N n η n = = n q=0 n q=0 [Φ q,n(η N q ) Φ q,n(φ q(η N q 1))] ( [Φ q,n Φ q(ηq 1) N + 1 ) ( ) ] Vn N Φ q,n Φ q(ηq 1) N N Semigroup/Stability/Uniform propagation of chaos/... dp+guionnet CRAS [99], and IHP [00], Miclo [00], [02], [03], Rousset [05],...

24 13/42 Another tool I N q := q-multi-indexes with values η (N,q) n (f ) := m(ξ n ) q := 1 (N) q c I N q δ ( ξ c(1) n ),...,ξn c(q)

25 13/42 Another tool I N q := q-multi-indexes with values η (N,q) n (f ) := m(ξ n ) q := 1 (N) q c I N q δ ( ξ c(1) n ),...,ξn c(q) combinatorial/transport/coalescent tree formulae between m(ξ n ) q and m(ξ n ) q and m(ξ n ) q m(ξ n ) q tv (q 1) 2 /N

26 13/42 Another tool I N q := q-multi-indexes with values η (N,q) n (f ) := m(ξ n ) q := 1 (N) q c I N q δ ( ξ c(1) n ),...,ξn c(q) combinatorial/transport/coalescent tree formulae between m(ξ n ) q and m(ξ n ) q and m(ξ n ) q m(ξ n ) q tv (q 1) 2 /N Why these objects?

27 13/42 Another tool I N q := q-multi-indexes with values η (N,q) n (f ) := m(ξ n ) q := 1 (N) q c I N q δ ( ξ c(1) n ),...,ξn c(q) combinatorial/transport/coalescent tree formulae between m(ξ n ) q and m(ξ n ) q and m(ξ n ) q m(ξ n ) q tv (q 1) 2 /N Why these objects? Propagation of chaos = Bias E ( f (ξn, 1..., ξn) q ) ( ) ξ n 1 = E η n (N,q) (f ) ξ n 1 = η (N,q) n 1 K q m(ξ (f ) n 1)

28 Another tool I N q := q-multi-indexes with values η (N,q) n (f ) := m(ξ n ) q := 1 (N) q c I N q δ ( ξ c(1) n ),...,ξn c(q) combinatorial/transport/coalescent tree formulae between m(ξ n ) q and m(ξ n ) q and m(ξ n ) q m(ξ n ) q tv (q 1) 2 /N Why these objects? Propagation of chaos = Bias E ( f (ξn, 1..., ξn) q ) ( ) ξ n 1 = E η n (N,q) (f ) ξ n 1 = η (N,q) n 1 K q m(ξ (f ) n 1) Non asymp. Taylor exp.+ Uniform propagation of chaos q-particles 13/42 dp Springer [04], dp+peters SIAM [06], dp+patras+rubenthaler AoAP [06], dp+miclo SAA[07] Chan+dp+Jasra Arxiv [14]

29 14/42 Nonlinear Markov chains Mean field particle methods Some stochastic perturbation tools Feynman-Kac models ( filtering, rare events,... ) A brief review on genetic type models Non commutative models Lyapunov weighted dynamics model Particle Gibbs-Glauber dynamics Continuous time models

30 15/42 Feynman-Kac measures with Q n (F n ) = Γ n (F n )/Γ n (1) and η n (f n ) = γ n (f n )/γ n (1) Γ n (F n ) = E (F n (X n ) Z n (X )) and γ n (f n ) = E (f n (X n ) Z n (X )) with the historical process X n = (X 0,..., X n ) and Z n (X ) = G k (X k ) 0 k<n

31 15/42 Feynman-Kac measures with Q n (F n ) = Γ n (F n )/Γ n (1) and η n (f n ) = γ n (f n )/γ n (1) Γ n (F n ) = E (F n (X n ) Z n (X )) and γ n (f n ) = E (f n (X n ) Z n (X )) with the historical process X n = (X 0,..., X n ) and Z n (X ) = G k (X k ) 0 k<n Mean field particle Genetic/Branching process ξ n = (ξ i n) i η n η N n = 1 δ N ξ i n 1 i N }{{} :=m(ξ n)

32 15/42 Feynman-Kac measures with Q n (F n ) = Γ n (F n )/Γ n (1) and η n (f n ) = γ n (f n )/γ n (1) Γ n (F n ) = E (F n (X n ) Z n (X )) and γ n (f n ) = E (f n (X n ) Z n (X )) with the historical process X n = (X 0,..., X n ) and Z n (X ) = G k (X k ) 0 k<n Mean field particle Genetic/Branching process ξ n = (ξ i n) i η n η N n = 1 δ N ξ i n 1 i N }{{} :=m(ξ n) path-space = 1 δ N (ξ i 0,n,ξ1,n i,...,ξi n,n ) 1 i N }{{} =Law ancestral line := X n Q n

33 16/42 Graphical illustration : η n η N n := 1 N 1 i N δ ξ i n

34 16/42 Graphical illustration : η n η N n := 1 N 1 i N δ ξ i n

35 16/42 Graphical illustration : η n η N n := 1 N 1 i N δ ξ i n

36 16/42 Graphical illustration : η n η N n := 1 N 1 i N δ ξ i n

37 16/42 Graphical illustration : η n η N n := 1 N 1 i N δ ξ i n

38 16/42 Graphical illustration : η n η N n := 1 N 1 i N δ ξ i n

39 16/42 Graphical illustration : η n η N n := 1 N 1 i N δ ξ i n

40 16/42 Graphical illustration : η n η N n := 1 N 1 i N δ ξ i n

41 16/42 Graphical illustration : η n η N n := 1 N 1 i N δ ξ i n

42 16/42 Graphical illustration : η n η N n := 1 N 1 i N δ ξ i n

43 16/42 Graphical illustration : η n η N n := 1 N 1 i N δ ξ i n

44 16/42 Graphical illustration : η n η N n := 1 N 1 i N δ ξ i n

45 16/42 Graphical illustration : η n η N n := 1 N 1 i N δ ξ i n

46 16/42 Graphical illustration : η n η N n := 1 N 1 i N δ ξ i n

47 16/42 Graphical illustration : η n η N n := 1 N 1 i N δ ξ i n

48 16/42 Graphical illustration : η n η N n := 1 N 1 i N δ ξ i n

49 16/42 Graphical illustration : η n η N n := 1 N 1 i N δ ξ i n

50 16/42 Graphical illustration : η n η N n := 1 N 1 i N δ ξ i n

51 16/42 Graphical illustration : η n η N n := 1 N 1 i N δ ξ i n

52 17/42 Equivalent particle algorithms Sequential Monte Carlo Sampling Resampling Particle Filters Prediction Updating Genetic Algorithms Mutation Selection Evolutionary Population Exploration Branching-selection Diffusion Monte Carlo Free evolutions Absorption Quantum Monte Carlo Walkers motions Reconfiguration Sampling Algorithms Transition proposals Accept-reject-recycle

53 17/42 Equivalent particle algorithms Sequential Monte Carlo Sampling Resampling Particle Filters Prediction Updating Genetic Algorithms Mutation Selection Evolutionary Population Exploration Branching-selection Diffusion Monte Carlo Free evolutions Absorption Quantum Monte Carlo Walkers motions Reconfiguration Sampling Algorithms Transition proposals Accept-reject-recycle Many other lively buzzwords : bootstrapping, spawning, cloning, pruning, replenish, splitting, enrichment, go with the winner, look-ahead, weighted dynamics, quantum teleportation, Heuristic style algo Particle Feynman-Kac models Convergence analysis : Uniform propagation of chaos, concentration, stability, CLT, LDP, MDP, Empirical processes,...

54 18/42 Feynman-Kac measures Unnormalized measures γ n (f n ) = η n (f n ) η p (G p ) 0 p<n

55 18/42 Feynman-Kac measures Unnormalized measures γ n (f n ) = η n (f n ) η p (G p ) 0 p<n Backward Markov chain formula G n 1 (x n 1 )M n (x n 1, dx n ) = H n (x n 1, x n ) λ n (dx n ) Q n (d(x 0,..., x n )) = η n (dx n ) M q+1,ηq (x q+1, dx q ) }{{} 0 q<n η q(dx q) H q+1(x q,x q+1)

56 18/42 Feynman-Kac measures Unnormalized measures γ n (f n ) = η n (f n ) η p (G p ) 0 p<n Backward Markov chain formula G n 1 (x n 1 )M n (x n 1, dx n ) = H n (x n 1, x n ) λ n (dx n ) Q n (d(x 0,..., x n )) = η n (dx n ) M q+1,ηq (x q+1, dx q ) }{{} 0 q<n η q(dx q) H q+1(x q,x q+1) Application domains : Any conditional distributions, Bayesian prior/posterior, nonlinear filtering, Boltzmann-Gibbs measures, branching processes, Twisted importance sampling,...

57 19/42 Non commutative models with G n (x n ) R d d s.t. u S d 1 := { u = 1} we have G n (x) u > 0 f n (x 0,..., x n ) R d and 0 p n A p = A 0 A 1... A n Γ n (f n ).u 0 := E f n (X 0,..., X n ) = E f n (X 0,..., X n ) 0 p<n 0 p<n G p (X p ).u 0 G p (X p ) X n = (X n, U n ) ( E n S d 1) and G n (X n ) = G n (X n ).U n and the walk on the sphere model U n+1 = G n(x n ).U n G n (X n ).U n

58 20/42 of P n (ϕ)(x) = E x (ϕ(y n )) with Y n+1 = F n (Y n, W n ) First variational equation Jac (Y n+1 ) = G n (Y n, W n ) Jac (Y n ) with Gn i,j = x j Fn i = G p (X p ) with X n = (Y n, W n ) 0 p n P n (ϕ)(x).u 0 = E f n (X n ) }{{} = (ϕ)(y n).u n 0 p<n G p (X p ) }{{} = G p(x p).u p FK model w.r.t. Y n weighted with the directional Lyap. exp. 0 p n G p (X p ) = Jac (Y n ).u 0 = 0 p n Jac (Y p ).u 0 Jac (Y p 1 ).u 0

59 1/42 Related Feynman-Kac model X n = (X n, X n+1 ) and G n (X n ) = Jac (X n+1 ) α / Jac (X n ) α Feynman-Kac model = The Lyapunov weighted dynamics model dq n = 1 Z n Jac (X n ) α dp n α < 0 dq n favors low Lyapunov trajectories α > 0 dq n favors high Lyapunov trajectories Hyper-references : T Laffargue, K.D. Nguyen Thu Lam, J. Kurchan, J. Tailleur LDP of Lyapunov exp. (2013) S. Tanese Nicola, J. Kurchan. Metastable states, transitions, basins and borders at finite temperature J. Stat. Phys. (2004). J. Tailleur, S. Tanese Nicola, J. Kurchan. Kramers equations an supersymmetry J. Stat. Phys. (2006). J. Tailleur, J. Kurchan. Probing rare physical trajectories with Lyapunov weighted dynamics, Nature Physics (2007) Genealogical particle analysis of aare events (joint work with J. Garnier)AAP (2005).

60 22/42 Particle Feynman-Kac measures Unbiased particle unnormalized measures (dp MPRF (96)) γn N (f n ) := ηn N (f n ) ηp N (G p ) 0 p<n

61 22/42 Particle Feynman-Kac measures Unbiased particle unnormalized measures (dp MPRF (96)) γn N (f n ) := ηn N (f n ) ηp N (G p ) 0 p<n Particle backward model (dp+doucet+singh HAL (09), M2AN (10)) Q N n (d(x 0,..., x n )) := ηn N (dx n ) M q+1,η N q (x q+1, dx q ) }{{} 0 q<n ηq N (dxq) Hq+1(xq,xq+1)

62 23/42 A first theorem Theorem [dp+kohn+patras, Arxiv (14), CRAS (15)+IHP (16)] Law ( X n (ξ k,k ) 0 k n ) = Law of backward ancestral line X n given all pop.

63 3/42 A first theorem Theorem [dp+kohn+patras, Arxiv (14), CRAS (15)+IHP (16)] Law ( X n (ξ k,k ) 0 k n ) = Law of backward ancestral line X n given all pop. := ηn N (dx n ) M n,η N n 1 (x n, dx n 1 ) M 1,η N 0 (x 1, dx 0 ) with the occupation measures η N k = m(ξ k,k ) := 1 N 1 i N δ ξ i k,k

64 24/42 Nonlinear Markov chains Mean field particle methods Some stochastic perturbation tools Feynman-Kac models ( filtering, rare events,... ) Particle Gibbs-Glauber dynamics Many body Feynman-Kac (particle) models Duality formulae + Taylor expansions Branching proc. and Interacting MCMC Continuous time models

65 25/42 Many body FK measures π n and F n (ξ n ) := m(ξ n )(f n ) and G n (ξ n ) := m(ξ n )(G n ) Z n (ξ) := G k (ξ k ) Z n = η k (G k ) 0 k<n 0 k<n

66 25/42 Many body FK measures π n and F n (ξ n ) := m(ξ n )(f n ) and G n (ξ n ) := m(ξ n )(G n ) Z n (ξ) := G k (ξ k ) Z n = η k (G k ) 0 k<n 0 k<n Unbiasedness property in terms of many-body FK measures π n (F n ) : E (F n (ξ n ) Z n (ξ))

67 5/42 Many body FK measures π n and F n (ξ n ) := m(ξ n )(f n ) and G n (ξ n ) := m(ξ n )(G n ) Z n (ξ) := G k (ξ k ) Z n = η k (G k ) 0 k<n 0 k<n Unbiasedness property in terms of many-body FK measures π n (F n ) : E (F n (ξ n ) Z n (ξ)) = E (f n (X n ) Z n (X )) η n (f n ) (Note: the definition of π n if for any symmetric functions F n (ξ n ) )

68 5/42 Many body FK measures π n and F n (ξ n ) := m(ξ n )(f n ) and G n (ξ n ) := m(ξ n )(G n ) Z n (ξ) := G k (ξ k ) Z n = η k (G k ) 0 k<n 0 k<n Unbiasedness property in terms of many-body FK measures π n (F n ) : E (F n (ξ n ) Z n (ξ)) = E (f n (X n ) Z n (X )) η n (f n ) (Note: the definition of π n if for any symmetric functions F n (ξ n ) ) Unbiasedness property in terms of ancestral lines ( ) E (f n (X n ) Z n (ξ)) = E f n (X n) Z n (ξ) = E (f n (X n ) Z n (X ))

69 26/42 Second Extended Many body FK measures π n Definition: (Extended) Many-body FK measures π n (F n ) : E (F n (X n, ξ n ) Z n (ξ)) with historical process ξ n = (ξ 0,..., ξ n ) (ancestral lines evolution).

70 26/42 Second Extended Many body FK measures π n Definition: (Extended) Many-body FK measures π n (F n ) : E (F n (X n, ξ n ) Z n (ξ)) with historical process ξ n = (ξ 0,..., ξ n ) (ancestral lines evolution). Particle Gibbs-Glauber dynamics (PGD) Target Law π (X n, ξ n )

71 26/42 Second Extended Many body FK measures π n Definition: (Extended) Many-body FK measures π n (F n ) : E (F n (X n, ξ n ) Z n (ξ)) with historical process ξ n = (ξ 0,..., ξ n ) (ancestral lines evolution). Particle Gibbs-Glauber dynamics (PGD) Target Law π (X n, ξ n ) } { } { X n = z Xn = z (X n ξ n = x) Xn = z ξ n = x ξ n = x ξ n = x (ξ n X n = z)

72 26/42 Second Extended Many body FK measures π n Definition: (Extended) Many-body FK measures π n (F n ) : E (F n (X n, ξ n ) Z n (ξ)) with historical process ξ n = (ξ 0,..., ξ n ) (ancestral lines evolution). Particle Gibbs-Glauber dynamics (PGD) Target Law π (X n, ξ n ) } { } { X n = z Xn = z (X n ξ n = x) Xn = z ξ n = x ξ n = x ξ n = x (ξ n X n = z) Important observation: X n X n is a Markov process on path space

73 Duality formula To sample the second PGD transition. 27/42

74 27/42 Duality formula To sample the second PGD transition. Theo.- Duality [dp+kohn+patras, Arxiv (14), CRAS (15)+IHP (16)] F n E (F n (X n, ξ n ) Z n (ξ)) = E ( F n (X n, ξ n ) Z n(x ) ) with the dual process: ξ n defined as ξ n but with a given frozen ancestral line X n

75 27/42 Duality formula To sample the second PGD transition. Theo.- Duality [dp+kohn+patras, Arxiv (14), CRAS (15)+IHP (16)] F n E (F n (X n, ξ n ) Z n (ξ)) = E ( F n (X n, ξ n ) Z n(x ) ) with the dual process: ξ n defined as ξ n but with a given frozen ancestral line X n Corollary - Backward duality Same duality PGD when (X n, ξ n ) is replaced by (X n, ξ n,n ) with the population historical process ξ n,n = (ξ k,k ) 0 k n

76 28/42 Taylor expansions X n X n Markov process on path space with η n -reversible transition K n n = FIXED length/duration of the ancestral trajectories N = number of particles used in the genealogical trees

77 28/42 Taylor expansions X n X n Markov process on path space with η n -reversible transition K n n = FIXED length/duration of the ancestral trajectories N = number of particles used in the genealogical trees Theo. [dp+kohn+patras, Arxiv (14), CRAS (15)+IHP (16)] K n (x,.) = η n + 1 k l 1 N k d (k) K n (x,.) + 1 N l+1 l+1 K n (x,.) ( l 1) Derivatives coalescent/colored trees.

78 Taylor expansions X n X n Markov process on path space with η n -reversible transition K n n = FIXED length/duration of the ancestral trajectories N = number of particles used in the genealogical trees Theo. [dp+kohn+patras, Arxiv (14), CRAS (15)+IHP (16)] K n (x,.) = η n + 1 k l 1 N k d (k) K n (x,.) + 1 N l+1 l+1 K n (x,.) ( l 1) Derivatives coalescent/colored trees. d (k) K n (cnk 2 ) k and (l+1) K n (cn(l + 1) 2 ) l+1 28/42 as soon as N > cn(l + 1) 2, for some c <.

79 A couple of direct corollaries 29/42

80 29/42 A couple of direct corollaries Cor. 1: m 1 and osc(f ) 1 K m n (f )(x) η n (f ) β (K m n ) (cn/n) m ( min (N,m) 0 )

81 29/42 A couple of direct corollaries Cor. 1: m 1 and osc(f ) 1 K m n (f )(x) η n (f ) β (K m n ) (cn/n) m ( min (N,m) 0 ) Cor. 2: m 1, p 1 and f 1 ] m(f Km n (f ) η n (f ) Lp(η n) [d N m (1) K n ) Lp(η n) (cn/n)m+1

82 30/42 Some comments Spatial branching processes varying pop. size N n First moments FK-models FK-particle = fixed pop.size

83 30/42 Some comments Spatial branching processes varying pop. size N n First moments FK-models FK-particle = fixed pop.size Particle/Absorption models = FK models with potential G [0, 1] η n = Law(X c n T killing > n) η = quasi-invariant meas.

84 30/42 Some comments Spatial branching processes varying pop. size N n First moments FK-models FK-particle = fixed pop.size Particle/Absorption models = FK models with potential G [0, 1] η n = Law(X c n T killing > n) η = quasi-invariant meas. Mutations = MCMC with target η n e βnv (x) λ(dx) and branching rates/selection weights e (βn+1 βn)v (x) β n = interacting MCMC/SMC/...

85 30/42 Some comments Spatial branching processes varying pop. size N n First moments FK-models FK-particle = fixed pop.size Particle/Absorption models = FK models with potential G [0, 1] η n = Law(X c n T killing > n) η = quasi-invariant meas. Mutations = MCMC with target η n e βnv (x) λ(dx) and branching rates/selection weights e (βn+1 βn)v (x) β n = interacting MCMC/SMC/... Mutations = MCMC with target η n 1 An (x) λ(dx) and branching rates/selection weights 1 An+1 A n = interacting MCMC/SMC/Multilevel splitting...

86 31/42 Nonlinear Markov chains Mean field particle methods Some stochastic perturbation tools Feynman-Kac models ( filtering, rare events,... ) Particle Gibbs-Glauber dynamics Continuous time models Feynman-Kac models - Diffusion Monte Carlo Gradient type diffusions Ensemble Kalman-Bucy filter Stability + Unif. prop. chaos

87 32/42 (weak sense) : infinitesimal generators L t,η t η t (f ) = η t L t,ηt (f ) := η t (dx) L t,ηt (f )(x) L t,η regular familly X t ( McKean measure) Ex. jump-diff. S L t,η (f )(x) = a t (x, η) f (x) σ2 t (x, η) f (x) +λ t (x, η) (f (y) f (x)) K t,η (x, dy)

88 32/42 (weak sense) : infinitesimal generators L t,η t η t (f ) = η t L t,ηt (f ) := η t (dx) L t,ηt (f )(x) L t,η regular familly X t ( McKean measure) Ex. jump-diff. S L t,η (f )(x) = a t (x, η) f (x) σ2 t (x, η) f (x) +λ t (x, η) (f (y) f (x)) K t,η (x, dy) Mean field particle model ξt i with generator L t,η N t with ηt N = N 1 1 i N δ ξ i t

89 2/42 (weak sense) : infinitesimal generators L t,η t η t (f ) = η t L t,ηt (f ) := η t (dx) L t,ηt (f )(x) L t,η regular familly X t ( McKean measure) Ex. jump-diff. S L t,η (f )(x) = a t (x, η) f (x) σ2 t (x, η) f (x) +λ t (x, η) (f (y) f (x)) K t,η (x, dy) Mean field particle model ξ i t with generator L t,η N t with η N t = N 1 Local perturbation eq. 1 i N dη N t (f ) = η N t L t,η N t (f ) dt + 1 N dm N t (f ) δ ξ i t

90 33/42 FKS-semigroups γ t = γ s Q s,t Normalization ( { t } ) Q s,t (f )(x) = E f (X t ) exp V u (X u )du X s = x s η t (f ) := γ t (f )/γ t (1) Jump-diffusion particles with L t,η (f )(x) = L X (f )(x) + V (x) (f (y) f (x)) η t (dy) Denormalization ( t ) γ t (f ) = η t (f ) exp η s (V s )ds 0

91 34/42 FKS-Stochastic analysis Particle estimations ( [ t ]) E f (X t ) exp W (X s )ds 0 [ t = η t (f ) exp unbias N 0 [ t ηt N (f ) exp 0 ] η s (W )ds ] ηs N (W )ds

92 34/42 FKS-Stochastic analysis Particle estimations ( [ t ]) E f (X t ) exp W (X s )ds 0 [ t = η t (f ) exp unbias N 0 [ t ηt N (f ) exp 0 ] η s (W )ds ] ηs N (W )ds Ground states of Schrodinger op. : ( DMC, QMC) (v.p. λ ground state h (L µ-reversible)) [ t ] lim N,t ηn t h dµ et exp ηs N (W )ds e λt 0

93 34/42 FKS-Stochastic analysis Particle estimations ( [ t ]) E f (X t ) exp W (X s )ds 0 [ t = η t (f ) exp unbias N 0 [ t ηt N (f ) exp 0 ] η s (W )ds ] ηs N (W )ds Ground states of Schrodinger op. : ( DMC, QMC) (v.p. λ ground state h (L µ-reversible)) [ t ] lim N,t ηn t h dµ et exp ηs N (W )ds e λt 0 Asymptotic theory discrete time models. Some refs: dp+miclo [00,02,03,07] Mathias Rousset, Tony Lelièvre, and Gabriel Stoltz MATHERIALS/MICMAC INRIA team.

94 35/42 Gradient diff. - Ex: V (x) := x 4 4 x 2 2 and F (x) := α x 2 2 dx t = [ V + F η t ] ( ) X t dt + σdbt Applications : micro-macro-dilute physical models N = N 2 = particles algorithm?

95 35/42 Gradient diff. - Ex: V (x) := x 4 4 x 2 2 and F (x) := α x 2 2 dx t = [ V + F η t ] ( ) X t dt + σdbt Applications : micro-macro-dilute physical models N = N 2 = particles algorithm? V convex pioneering work of Malrieux [01]+Bakry-Emery tools.

96 35/42 Gradient diff. - Ex: V (x) := x 4 4 x 2 2 and F (x) := α x 2 2 dx t = [ V + F η t ] ( ) X t dt + σdbt Applications : micro-macro-dilute physical models N = N 2 = particles algorithm? V convex pioneering work of Malrieux [01]+Bakry-Emery tools. V and F symmetric + constant first moment (a.k.a. Law(X 0 ) symmetric or stationnary or? ) and α := inf F > sup V > 0 = F (x) := F (x) α x 2 2 and V (x) := V (x) + α x 2 2 convex case Carillo-McCann-Villani [03]

97 35/42 Gradient diff. - Ex: V (x) := x 4 4 x 2 2 and F (x) := α x 2 2 dx t = [ V + F η t ] ( ) X t dt + σdbt Applications : micro-macro-dilute physical models N = N 2 = particles algorithm? V convex pioneering work of Malrieux [01]+Bakry-Emery tools. V and F symmetric + constant first moment (a.k.a. Law(X 0 ) symmetric or stationnary or? ) and α := inf F > sup V > 0 = F (x) := F (x) α x 2 2 and V (x) := V (x) + α x 2 2 convex case Carillo-McCann-Villani [03] V non convex and σ too small non uniqueness invariant measures pionnering work by Tugaut [10], σ large enough and α > 1 Uniform propagation of chaos dp+tugaut [13].

98 36/42 Kalman-Bucy filter Linear+Gaussian filtering problem { dxt = A X t dt + R 1/2 dw t dy t = C X t dt + Σ 1/2 dv t F t := σ(y s, s t). Optimal L 2 -filter = Kalman-Bucy filter X t := E(X t F t ) and P t := E ( (X t E(X t F t )) (X t E(X t F t )) )

99 36/42 Kalman-Bucy filter Linear+Gaussian filtering problem { dxt = A X t dt + R 1/2 dw t dy t = C X t dt + Σ 1/2 dv t F t := σ(y s, s t). Optimal L 2 -filter = Kalman-Bucy filter X t := E(X t F t ) and P t := E ( (X t E(X t F t )) (X t E(X t F t )) ) d X t = A X t dt + P t C Σ 1 ( dy t C X ) t dt with the Riccati equation t P t = Ricc(P t ) := AP t + P t A P t SP t + R with S := C ΣC

100 37/42 Nonlinear Kalman-Bucy diffusion [ )] dx t = A X t dt + R 1/2 dw t +P ηt C Σ 1 dy t (CX t dt + Σ 1/2 dv t with the covariance matrices P ηt = η t [(e η t (e))(e η t (e)) ] with e(x) := x. E ( X t F t ) = Xt and P ηt = P t. Ensemble Kalman-Bucy filter = Interacting particles with empirical covariance p t := P η N t P t and m t = η N t (e) E(X t F t )

101 38/42 Nonlinear models Extended Kalman-Bucy-filters d X t = A( X t ) dt + P t C Σ 1 [ dy t C X ] t dt with the stochastic Riccati equation: t P t = A( X t )P t + P t A( X t ) + R P t SP t

102 38/42 Nonlinear models Extended Kalman-Bucy-filters d X t = A( X t ) dt + P t C Σ 1 [ dy t C X ] t dt with the stochastic Riccati equation: t P t = A( X t )P t + P t A( X t ) + R P t SP t McKean-Vlasov interpretation dx t = A ( X t, E[X t G t ] ) dt + R 1/2 dw t [ )] +P ηt C R 1 2 dy t (CX t dt + Σ 1/2 dv t with the drift function A(x, m) := A [m] + A [m] (x m).

103 39/42 Some illustrations Langevin type signal processes R = σ 2 Id and (A, A) = ( V, 2 V) Non quadratic potential (q R r, Q 1, Q 2 0) V(x) = 1 2 Q 1x, x + q, x Q 2x, x 3/2 Interacting diffusion gradient flows V(x) = U 1 (x i ) + 1 i r 1 i j r U 2 (x i, x j ) for some convex confining potential U i : R i [0, [

104 40/42 Regularity conditions Full observation S = s Id and λ A := sup x R r 1 λ max ( A(x) + A(x) ) < 0 A(x) A(y) κ A x y Examples: Langevin signal-diffusion ( ) (λ A, κ A ) = β 2 1 λ min (Q 1 ), 2λ 3/2 max(q 2 ). more generally 2 V v Id Lipschitz condition

105 41/42 Stability theorem (X t, Z t ) = McKean-Vlasov starting at (X 0, Z 0 ) Theo [dp-kurtzmann-tugaut] When λ A is sufficiently large we have W 2 (Law(X t ), Law(Z t )) c exp [ t λ] for some λ > 0. more explicit description in terms of (R, S, κ A ).

106 2/42 Propagation of chaos and P N t := Law(m t, p t ) P t := Law( X t, P t ) Q N t := Law(ξ 1 t ) Q t := Law(X t ) Theo SIAM Control & Opt [16] [dp-kurtzmann-tugaut] When λ A is sufficiently large, β ]0, 1/2] s.t. sup t 0 ( ) ( ) W 2 P N t, P t sup W 2 Q N t, Q t c N β t 0 Linear models unif. propagation of chaos with β = 1/2 for any stable drift matrix. Arxiv 1 (Stability KBF)+ A.N. Bishop Arxiv 2 (Perturbation KBF)+ A.N. Bishop Arxiv 3 (Unif. EnKBF)+ J. Tugaut. Arxiv 4 (Stability EKBF)+ A. Kurtzmann, J. Tugaut.

Mean field simulation for Monte Carlo integration. Part II : Feynman-Kac models. P. Del Moral

Mean field simulation for Monte Carlo integration. Part II : Feynman-Kac models. P. Del Moral Mean field simulation for Monte Carlo integration Part II : Feynman-Kac models P. Del Moral INRIA Bordeaux & Inst. Maths. Bordeaux & CMAP Polytechnique Lectures, INLN CNRS & Nice Sophia Antipolis Univ.

More information

A Backward Particle Interpretation of Feynman-Kac Formulae

A Backward Particle Interpretation of Feynman-Kac Formulae A Backward Particle Interpretation of Feynman-Kac Formulae P. Del Moral Centre INRIA de Bordeaux - Sud Ouest Workshop on Filtering, Cambridge Univ., June 14-15th 2010 Preprints (with hyperlinks), joint

More information

Concentration inequalities for Feynman-Kac particle models. P. Del Moral. INRIA Bordeaux & IMB & CMAP X. Journées MAS 2012, SMAI Clermond-Ferrand

Concentration inequalities for Feynman-Kac particle models. P. Del Moral. INRIA Bordeaux & IMB & CMAP X. Journées MAS 2012, SMAI Clermond-Ferrand Concentration inequalities for Feynman-Kac particle models P. Del Moral INRIA Bordeaux & IMB & CMAP X Journées MAS 2012, SMAI Clermond-Ferrand Some hyper-refs Feynman-Kac formulae, Genealogical & Interacting

More information

A new class of interacting Markov Chain Monte Carlo methods

A new class of interacting Markov Chain Monte Carlo methods A new class of interacting Marov Chain Monte Carlo methods P Del Moral, A Doucet INRIA Bordeaux & UBC Vancouver Worshop on Numerics and Stochastics, Helsini, August 2008 Outline 1 Introduction Stochastic

More information

Advanced Monte Carlo integration methods. P. Del Moral (INRIA team ALEA) INRIA & Bordeaux Mathematical Institute & X CMAP

Advanced Monte Carlo integration methods. P. Del Moral (INRIA team ALEA) INRIA & Bordeaux Mathematical Institute & X CMAP Advanced Monte Carlo integration methods P. Del Moral (INRIA team ALEA) INRIA & Bordeaux Mathematical Institute & X CMAP MCQMC 2012, Sydney, Sunday Tutorial 12-th 2012 Some hyper-refs Feynman-Kac formulae,

More information

An introduction to particle simulation of rare events

An introduction to particle simulation of rare events An introduction to particle simulation of rare events P. Del Moral Centre INRIA de Bordeaux - Sud Ouest Workshop OPUS, 29 juin 2010, IHP, Paris P. Del Moral (INRIA) INRIA Bordeaux-Sud Ouest 1 / 33 Outline

More information

Mean field simulation for Monte Carlo integration. Part I : Intro. nonlinear Markov models (+links integro-diff eq.) P. Del Moral

Mean field simulation for Monte Carlo integration. Part I : Intro. nonlinear Markov models (+links integro-diff eq.) P. Del Moral Mean field simulation for Monte Carlo integration Part I : Intro. nonlinear Markov models (+links integro-diff eq.) P. Del Moral INRIA Bordeaux & Inst. Maths. Bordeaux & CMAP Polytechnique Lectures, INLN

More information

Contraction properties of Feynman-Kac semigroups

Contraction properties of Feynman-Kac semigroups Journées de Statistique Marne La Vallée, January 2005 Contraction properties of Feynman-Kac semigroups Pierre DEL MORAL, Laurent MICLO Lab. J. Dieudonné, Nice Univ., LATP Univ. Provence, Marseille 1 Notations

More information

On the Concentration Properties of Interacting Particle Processes. Contents

On the Concentration Properties of Interacting Particle Processes. Contents Foundations and Trends R in Machine Learning Vol. 3, Nos. 3 4 (2010) 225 389 c 2012 P. Del Moral, P. Hu, and L. Wu DOI: 10.1561/2200000026 On the Concentration Properties of Interacting Particle Processes

More information

Sequential Monte Carlo Methods for Bayesian Computation

Sequential Monte Carlo Methods for Bayesian Computation Sequential Monte Carlo Methods for Bayesian Computation A. Doucet Kyoto Sept. 2012 A. Doucet (MLSS Sept. 2012) Sept. 2012 1 / 136 Motivating Example 1: Generic Bayesian Model Let X be a vector parameter

More information

Full text available at: On the Concentration Properties of Interacting Particle Processes

Full text available at:  On the Concentration Properties of Interacting Particle Processes On the Concentration Properties of Interacting Particle Processes On the Concentration Properties of Interacting Particle Processes Pierre Del Moral INRIA, Université Bordeaux 1 France Peng Hu INRIA, Université

More information

Importance splitting for rare event simulation

Importance splitting for rare event simulation Importance splitting for rare event simulation F. Cerou Frederic.Cerou@inria.fr Inria Rennes Bretagne Atlantique Simulation of hybrid dynamical systems and applications to molecular dynamics September

More information

Stability of Feynman-Kac Semigroup and SMC Parameters' Tuning

Stability of Feynman-Kac Semigroup and SMC Parameters' Tuning Stability of Feynman-Kac Semigroup and SMC Parameters' Tuning François GIRAUD PhD student, CEA-CESTA, INRIA Bordeaux Tuesday, February 14th 2012 1 Recalls and context of the analysis Feynman-Kac semigroup

More information

A Central Limit Theorem for Fleming-Viot Particle Systems Application to the Adaptive Multilevel Splitting Algorithm

A Central Limit Theorem for Fleming-Viot Particle Systems Application to the Adaptive Multilevel Splitting Algorithm A Central Limit Theorem for Fleming-Viot Particle Systems Application to the Algorithm F. Cérou 1,2 B. Delyon 2 A. Guyader 3 M. Rousset 1,2 1 Inria Rennes Bretagne Atlantique 2 IRMAR, Université de Rennes

More information

On the concentration properties of Interacting particle processes

On the concentration properties of Interacting particle processes ISTITUT ATIOAL DE RECHERCHE E IFORMATIQUE ET E AUTOMATIQUE arxiv:1107.1948v2 [math.a] 12 Jul 2011 On the concentration properties of Interacting particle processes Pierre Del Moral, Peng Hu, Liming Wu

More information

A Backward Particle Interpretation of Feynman-Kac Formulae

A Backward Particle Interpretation of Feynman-Kac Formulae A Backward Particle Interpretation of Feynman-Kac Formulae Pierre Del Moral, Arnaud Doucet, Sumeetpal Singh To cite this version: Pierre Del Moral, Arnaud Doucet, Sumeetpal Singh. A Backward Particle Interpretation

More information

Numerical methods in molecular dynamics and multiscale problems

Numerical methods in molecular dynamics and multiscale problems Numerical methods in molecular dynamics and multiscale problems Two examples T. Lelièvre CERMICS - Ecole des Ponts ParisTech & MicMac project-team - INRIA Horizon Maths December 2012 Introduction The aim

More information

Effective dynamics for the (overdamped) Langevin equation

Effective dynamics for the (overdamped) Langevin equation Effective dynamics for the (overdamped) Langevin equation Frédéric Legoll ENPC and INRIA joint work with T. Lelièvre (ENPC and INRIA) Enumath conference, MS Numerical methods for molecular dynamics EnuMath

More information

Computer Intensive Methods in Mathematical Statistics

Computer Intensive Methods in Mathematical Statistics Computer Intensive Methods in Mathematical Statistics Department of mathematics johawes@kth.se Lecture 7 Sequential Monte Carlo methods III 7 April 2017 Computer Intensive Methods (1) Plan of today s lecture

More information

Data assimilation as an optimal control problem and applications to UQ

Data assimilation as an optimal control problem and applications to UQ Data assimilation as an optimal control problem and applications to UQ Walter Acevedo, Angwenyi David, Jana de Wiljes & Sebastian Reich Universität Potsdam/ University of Reading IPAM, November 13th 2017

More information

Bridging the Gap between Center and Tail for Multiscale Processes

Bridging the Gap between Center and Tail for Multiscale Processes Bridging the Gap between Center and Tail for Multiscale Processes Matthew R. Morse Department of Mathematics and Statistics Boston University BU-Keio 2016, August 16 Matthew R. Morse (BU) Moderate Deviations

More information

ECE276A: Sensing & Estimation in Robotics Lecture 10: Gaussian Mixture and Particle Filtering

ECE276A: Sensing & Estimation in Robotics Lecture 10: Gaussian Mixture and Particle Filtering ECE276A: Sensing & Estimation in Robotics Lecture 10: Gaussian Mixture and Particle Filtering Lecturer: Nikolay Atanasov: natanasov@ucsd.edu Teaching Assistants: Siwei Guo: s9guo@eng.ucsd.edu Anwesan Pal:

More information

An Brief Overview of Particle Filtering

An Brief Overview of Particle Filtering 1 An Brief Overview of Particle Filtering Adam M. Johansen a.m.johansen@warwick.ac.uk www2.warwick.ac.uk/fac/sci/statistics/staff/academic/johansen/talks/ May 11th, 2010 Warwick University Centre for Systems

More information

On a class of stochastic differential equations in a financial network model

On a class of stochastic differential equations in a financial network model 1 On a class of stochastic differential equations in a financial network model Tomoyuki Ichiba Department of Statistics & Applied Probability, Center for Financial Mathematics and Actuarial Research, University

More information

EnKF-based particle filters

EnKF-based particle filters EnKF-based particle filters Jana de Wiljes, Sebastian Reich, Wilhelm Stannat, Walter Acevedo June 20, 2017 Filtering Problem Signal dx t = f (X t )dt + 2CdW t Observations dy t = h(x t )dt + R 1/2 dv t.

More information

Perfect simulation algorithm of a trajectory under a Feynman-Kac law

Perfect simulation algorithm of a trajectory under a Feynman-Kac law Perfect simulation algorithm of a trajectory under a Feynman-Kac law Data Assimilation, 24th-28th September 212, Oxford-Man Institute C. Andrieu, N. Chopin, A. Doucet, S. Rubenthaler University of Bristol,

More information

Multilevel Monte Carlo for Stochastic McKean-Vlasov Equations

Multilevel Monte Carlo for Stochastic McKean-Vlasov Equations Multilevel Monte Carlo for Stochastic McKean-Vlasov Equations Lukasz Szpruch School of Mathemtics University of Edinburgh joint work with Shuren Tan and Alvin Tse (Edinburgh) Lukasz Szpruch (University

More information

An algebraic approach to stochastic duality. Cristian Giardinà

An algebraic approach to stochastic duality. Cristian Giardinà An algebraic approach to stochastic duality Cristian Giardinà RAQIS18, Annecy 14 September 2018 Collaboration with Gioia Carinci (Delft) Chiara Franceschini (Modena) Claudio Giberti (Modena) Jorge Kurchan

More information

Auxiliary Particle Methods

Auxiliary Particle Methods Auxiliary Particle Methods Perspectives & Applications Adam M. Johansen 1 adam.johansen@bristol.ac.uk Oxford University Man Institute 29th May 2008 1 Collaborators include: Arnaud Doucet, Nick Whiteley

More information

Lecture 6: Bayesian Inference in SDE Models

Lecture 6: Bayesian Inference in SDE Models Lecture 6: Bayesian Inference in SDE Models Bayesian Filtering and Smoothing Point of View Simo Särkkä Aalto University Simo Särkkä (Aalto) Lecture 6: Bayesian Inference in SDEs 1 / 45 Contents 1 SDEs

More information

Controlled sequential Monte Carlo

Controlled sequential Monte Carlo Controlled sequential Monte Carlo Jeremy Heng, Department of Statistics, Harvard University Joint work with Adrian Bishop (UTS, CSIRO), George Deligiannidis & Arnaud Doucet (Oxford) Bayesian Computation

More information

Improved diffusion Monte Carlo

Improved diffusion Monte Carlo Improved diffusion Monte Carlo Jonathan Weare University of Chicago with Martin Hairer (U of Warwick) October 4, 2012 Diffusion Monte Carlo The original motivation for DMC was to compute averages with

More information

2012 NCTS Workshop on Dynamical Systems

2012 NCTS Workshop on Dynamical Systems Barbara Gentz gentz@math.uni-bielefeld.de http://www.math.uni-bielefeld.de/ gentz 2012 NCTS Workshop on Dynamical Systems National Center for Theoretical Sciences, National Tsing-Hua University Hsinchu,

More information

Weakly interacting particle systems on graphs: from dense to sparse

Weakly interacting particle systems on graphs: from dense to sparse Weakly interacting particle systems on graphs: from dense to sparse Ruoyu Wu University of Michigan (Based on joint works with Daniel Lacker and Kavita Ramanan) USC Math Finance Colloquium October 29,

More information

Ergodicity in data assimilation methods

Ergodicity in data assimilation methods Ergodicity in data assimilation methods David Kelly Andy Majda Xin Tong Courant Institute New York University New York NY www.dtbkelly.com April 15, 2016 ETH Zurich David Kelly (CIMS) Data assimilation

More information

Introduction. log p θ (y k y 1:k 1 ), k=1

Introduction. log p θ (y k y 1:k 1 ), k=1 ESAIM: PROCEEDINGS, September 2007, Vol.19, 115-120 Christophe Andrieu & Dan Crisan, Editors DOI: 10.1051/proc:071915 PARTICLE FILTER-BASED APPROXIMATE MAXIMUM LIKELIHOOD INFERENCE ASYMPTOTICS IN STATE-SPACE

More information

Handbook of Stochastic Methods

Handbook of Stochastic Methods C. W. Gardiner Handbook of Stochastic Methods for Physics, Chemistry and the Natural Sciences Third Edition With 30 Figures Springer Contents 1. A Historical Introduction 1 1.1 Motivation I 1.2 Some Historical

More information

Homogenization with stochastic differential equations

Homogenization with stochastic differential equations Homogenization with stochastic differential equations Scott Hottovy shottovy@math.arizona.edu University of Arizona Program in Applied Mathematics October 12, 2011 Modeling with SDE Use SDE to model system

More information

Monte-Carlo Methods and Stochastic Processes

Monte-Carlo Methods and Stochastic Processes Monte-Carlo Methods and Stochastic Processes From Linear to Non-Linear EMMANUEL GOBET ECOLE POLYTECHNIQUE - UNIVERSITY PARIS-SACLAY CMAP, PALAISEAU CEDEX, FRANCE CRC Press Taylor & Francis Group 6000 Broken

More information

Markov Chain Monte Carlo Methods for Stochastic

Markov Chain Monte Carlo Methods for Stochastic Markov Chain Monte Carlo Methods for Stochastic Optimization i John R. Birge The University of Chicago Booth School of Business Joint work with Nicholas Polson, Chicago Booth. JRBirge U Florida, Nov 2013

More information

STATIONARY NON EQULIBRIUM STATES F STATES FROM A MICROSCOPIC AND A MACROSCOPIC POINT OF VIEW

STATIONARY NON EQULIBRIUM STATES F STATES FROM A MICROSCOPIC AND A MACROSCOPIC POINT OF VIEW STATIONARY NON EQULIBRIUM STATES FROM A MICROSCOPIC AND A MACROSCOPIC POINT OF VIEW University of L Aquila 1 July 2014 GGI Firenze References L. Bertini; A. De Sole; D. G. ; G. Jona-Lasinio; C. Landim

More information

STABILITY AND UNIFORM APPROXIMATION OF NONLINEAR FILTERS USING THE HILBERT METRIC, AND APPLICATION TO PARTICLE FILTERS 1

STABILITY AND UNIFORM APPROXIMATION OF NONLINEAR FILTERS USING THE HILBERT METRIC, AND APPLICATION TO PARTICLE FILTERS 1 The Annals of Applied Probability 0000, Vol. 00, No. 00, 000 000 STABILITY AND UNIFORM APPROXIMATION OF NONLINAR FILTRS USING TH HILBRT MTRIC, AND APPLICATION TO PARTICL FILTRS 1 By François LeGland and

More information

The tree-valued Fleming-Viot process with mutation and selection

The tree-valued Fleming-Viot process with mutation and selection The tree-valued Fleming-Viot process with mutation and selection Peter Pfaffelhuber University of Freiburg Joint work with Andrej Depperschmidt and Andreas Greven Population genetic models Populations

More information

Markov Chain Monte Carlo Methods for Stochastic Optimization

Markov Chain Monte Carlo Methods for Stochastic Optimization Markov Chain Monte Carlo Methods for Stochastic Optimization John R. Birge The University of Chicago Booth School of Business Joint work with Nicholas Polson, Chicago Booth. JRBirge U of Toronto, MIE,

More information

A Class of Fractional Stochastic Differential Equations

A Class of Fractional Stochastic Differential Equations Vietnam Journal of Mathematics 36:38) 71 79 Vietnam Journal of MATHEMATICS VAST 8 A Class of Fractional Stochastic Differential Equations Nguyen Tien Dung Department of Mathematics, Vietnam National University,

More information

Particle Filtering Approaches for Dynamic Stochastic Optimization

Particle Filtering Approaches for Dynamic Stochastic Optimization Particle Filtering Approaches for Dynamic Stochastic Optimization John R. Birge The University of Chicago Booth School of Business Joint work with Nicholas Polson, Chicago Booth. JRBirge I-Sim Workshop,

More information

Monte Carlo methods for sampling-based Stochastic Optimization

Monte Carlo methods for sampling-based Stochastic Optimization Monte Carlo methods for sampling-based Stochastic Optimization Gersende FORT LTCI CNRS & Telecom ParisTech Paris, France Joint works with B. Jourdain, T. Lelièvre, G. Stoltz from ENPC and E. Kuhn from

More information

Uniformly Uniformly-ergodic Markov chains and BSDEs

Uniformly Uniformly-ergodic Markov chains and BSDEs Uniformly Uniformly-ergodic Markov chains and BSDEs Samuel N. Cohen Mathematical Institute, University of Oxford (Based on joint work with Ying Hu, Robert Elliott, Lukas Szpruch) Centre Henri Lebesgue,

More information

Introduction to self-similar growth-fragmentations

Introduction to self-similar growth-fragmentations Introduction to self-similar growth-fragmentations Quan Shi CIMAT, 11-15 December, 2017 Quan Shi Growth-Fragmentations CIMAT, 11-15 December, 2017 1 / 34 Literature Jean Bertoin, Compensated fragmentation

More information

Lecture Particle Filters

Lecture Particle Filters FMS161/MASM18 Financial Statistics November 29, 2010 Monte Carlo filters The filter recursions could only be solved for HMMs and for linear, Gaussian models. Idea: Approximate any model with a HMM. Replace

More information

Variantes Algorithmiques et Justifications Théoriques

Variantes Algorithmiques et Justifications Théoriques complément scientifique École Doctorale MATISSE IRISA et INRIA, salle Markov jeudi 26 janvier 2012 Variantes Algorithmiques et Justifications Théoriques François Le Gland INRIA Rennes et IRMAR http://www.irisa.fr/aspi/legland/ed-matisse/

More information

Interacting and Annealing Particle Filters: Mathematics and a Recipe for Applications. Christoph Schnörr Bodo Rosenhahn Hans-Peter Seidel

Interacting and Annealing Particle Filters: Mathematics and a Recipe for Applications. Christoph Schnörr Bodo Rosenhahn Hans-Peter Seidel Interacting and Annealing Particle Filters: Mathematics and a Recipe for Applications Jürgen Gall Jürgen Potthoff Christoph Schnörr Bodo Rosenhahn Hans-Peter Seidel MPI I 2006 4 009 September 2006 Authors

More information

Robert Collins CSE586, PSU Intro to Sampling Methods

Robert Collins CSE586, PSU Intro to Sampling Methods Intro to Sampling Methods CSE586 Computer Vision II Penn State Univ Topics to be Covered Monte Carlo Integration Sampling and Expected Values Inverse Transform Sampling (CDF) Ancestral Sampling Rejection

More information

Lecture 6: Multiple Model Filtering, Particle Filtering and Other Approximations

Lecture 6: Multiple Model Filtering, Particle Filtering and Other Approximations Lecture 6: Multiple Model Filtering, Particle Filtering and Other Approximations Department of Biomedical Engineering and Computational Science Aalto University April 28, 2010 Contents 1 Multiple Model

More information

Exact Simulation of Diffusions and Jump Diffusions

Exact Simulation of Diffusions and Jump Diffusions Exact Simulation of Diffusions and Jump Diffusions A work by: Prof. Gareth O. Roberts Dr. Alexandros Beskos Dr. Omiros Papaspiliopoulos Dr. Bruno Casella 28 th May, 2008 Content 1 Exact Algorithm Construction

More information

l approche des grandes déviations

l approche des grandes déviations Épuisement de ressources partagées en environnement Markovien : l approche des grandes déviations Joint work with François Delarue (Paris Diderot) et René Schott (Nancy 1) Journées de Probabilité à Lille

More information

Introduction to Machine Learning CMU-10701

Introduction to Machine Learning CMU-10701 Introduction to Machine Learning CMU-10701 Markov Chain Monte Carlo Methods Barnabás Póczos & Aarti Singh Contents Markov Chain Monte Carlo Methods Goal & Motivation Sampling Rejection Importance Markov

More information

Advanced Computational Methods in Statistics: Lecture 5 Sequential Monte Carlo/Particle Filtering

Advanced Computational Methods in Statistics: Lecture 5 Sequential Monte Carlo/Particle Filtering Advanced Computational Methods in Statistics: Lecture 5 Sequential Monte Carlo/Particle Filtering Axel Gandy Department of Mathematics Imperial College London http://www2.imperial.ac.uk/~agandy London

More information

PROBABILITY: LIMIT THEOREMS II, SPRING HOMEWORK PROBLEMS

PROBABILITY: LIMIT THEOREMS II, SPRING HOMEWORK PROBLEMS PROBABILITY: LIMIT THEOREMS II, SPRING 218. HOMEWORK PROBLEMS PROF. YURI BAKHTIN Instructions. You are allowed to work on solutions in groups, but you are required to write up solutions on your own. Please

More information

Uncertainty in Kalman Bucy Filtering

Uncertainty in Kalman Bucy Filtering Uncertainty in Kalman Bucy Filtering Samuel N. Cohen Mathematical Institute University of Oxford Research Supported by: The Oxford Man Institute for Quantitative Finance The Oxford Nie Financial Big Data

More information

Mean-Field optimization problems and non-anticipative optimal transport. Beatrice Acciaio

Mean-Field optimization problems and non-anticipative optimal transport. Beatrice Acciaio Mean-Field optimization problems and non-anticipative optimal transport Beatrice Acciaio London School of Economics based on ongoing projects with J. Backhoff, R. Carmona and P. Wang Robust Methods in

More information

Stochastic Demography, Coalescents, and Effective Population Size

Stochastic Demography, Coalescents, and Effective Population Size Demography Stochastic Demography, Coalescents, and Effective Population Size Steve Krone University of Idaho Department of Mathematics & IBEST Demographic effects bottlenecks, expansion, fluctuating population

More information

Lecture 2: From Linear Regression to Kalman Filter and Beyond

Lecture 2: From Linear Regression to Kalman Filter and Beyond Lecture 2: From Linear Regression to Kalman Filter and Beyond January 18, 2017 Contents 1 Batch and Recursive Estimation 2 Towards Bayesian Filtering 3 Kalman Filter and Bayesian Filtering and Smoothing

More information

Convergence to equilibrium for rough differential equations

Convergence to equilibrium for rough differential equations Convergence to equilibrium for rough differential equations Samy Tindel Purdue University Barcelona GSE Summer Forum 2017 Joint work with Aurélien Deya (Nancy) and Fabien Panloup (Angers) Samy T. (Purdue)

More information

Markov Chain BSDEs and risk averse networks

Markov Chain BSDEs and risk averse networks Markov Chain BSDEs and risk averse networks Samuel N. Cohen Mathematical Institute, University of Oxford (Based on joint work with Ying Hu, Robert Elliott, Lukas Szpruch) 2nd Young Researchers in BSDEs

More information

Boltzmann-Gibbs Preserving Langevin Integrators

Boltzmann-Gibbs Preserving Langevin Integrators Boltzmann-Gibbs Preserving Langevin Integrators Nawaf Bou-Rabee Applied and Comp. Math., Caltech INI Workshop on Markov-Chain Monte-Carlo Methods March 28, 2008 4000 atom cluster simulation Governing Equations

More information

A Central Limit Theorem for Fleming-Viot Particle Systems 1

A Central Limit Theorem for Fleming-Viot Particle Systems 1 A Central Limit Theorem for Fleming-Viot Particle Systems 1 Frédéric Cérou 2 IRIA Rennes & IRMAR, France frederic.cerou@inria.fr Bernard Delyon Université Rennes 1 & IRMAR, France bernard.delyon@univ-rennes1.fr

More information

Convergence of Particle Filtering Method for Nonlinear Estimation of Vortex Dynamics

Convergence of Particle Filtering Method for Nonlinear Estimation of Vortex Dynamics Convergence of Particle Filtering Method for Nonlinear Estimation of Vortex Dynamics Meng Xu Department of Mathematics University of Wyoming February 20, 2010 Outline 1 Nonlinear Filtering Stochastic Vortex

More information

A NEW NONLINEAR FILTER

A NEW NONLINEAR FILTER COMMUNICATIONS IN INFORMATION AND SYSTEMS c 006 International Press Vol 6, No 3, pp 03-0, 006 004 A NEW NONLINEAR FILTER ROBERT J ELLIOTT AND SIMON HAYKIN Abstract A discrete time filter is constructed

More information

Lecture Particle Filters. Magnus Wiktorsson

Lecture Particle Filters. Magnus Wiktorsson Lecture Particle Filters Magnus Wiktorsson Monte Carlo filters The filter recursions could only be solved for HMMs and for linear, Gaussian models. Idea: Approximate any model with a HMM. Replace p(x)

More information

CPSC 540: Machine Learning

CPSC 540: Machine Learning CPSC 540: Machine Learning MCMC and Non-Parametric Bayes Mark Schmidt University of British Columbia Winter 2016 Admin I went through project proposals: Some of you got a message on Piazza. No news is

More information

Sequential Monte Carlo Methods in High Dimensions

Sequential Monte Carlo Methods in High Dimensions Sequential Monte Carlo Methods in High Dimensions Alexandros Beskos Statistical Science, UCL Oxford, 24th September 2012 Joint work with: Dan Crisan, Ajay Jasra, Nik Kantas, Andrew Stuart Imperial College,

More information

Intertwinings for Markov processes

Intertwinings for Markov processes Intertwinings for Markov processes Aldéric Joulin - University of Toulouse Joint work with : Michel Bonnefont - Univ. Bordeaux Workshop 2 Piecewise Deterministic Markov Processes ennes - May 15-17, 2013

More information

Mathematical Methods for Neurosciences. ENS - Master MVA Paris 6 - Master Maths-Bio ( )

Mathematical Methods for Neurosciences. ENS - Master MVA Paris 6 - Master Maths-Bio ( ) Mathematical Methods for Neurosciences. ENS - Master MVA Paris 6 - Master Maths-Bio (2014-2015) Etienne Tanré - Olivier Faugeras INRIA - Team Tosca November 26th, 2014 E. Tanré (INRIA - Team Tosca) Mathematical

More information

Robert Collins CSE586, PSU Intro to Sampling Methods

Robert Collins CSE586, PSU Intro to Sampling Methods Intro to Sampling Methods CSE586 Computer Vision II Penn State Univ Topics to be Covered Monte Carlo Integration Sampling and Expected Values Inverse Transform Sampling (CDF) Ancestral Sampling Rejection

More information

Part 1: Expectation Propagation

Part 1: Expectation Propagation Chalmers Machine Learning Summer School Approximate message passing and biomedicine Part 1: Expectation Propagation Tom Heskes Machine Learning Group, Institute for Computing and Information Sciences Radboud

More information

Path Decomposition of Markov Processes. Götz Kersting. University of Frankfurt/Main

Path Decomposition of Markov Processes. Götz Kersting. University of Frankfurt/Main Path Decomposition of Markov Processes Götz Kersting University of Frankfurt/Main joint work with Kaya Memisoglu, Jim Pitman 1 A Brownian path with positive drift 50 40 30 20 10 0 0 200 400 600 800 1000-10

More information

Semicircle law on short scales and delocalization for Wigner random matrices

Semicircle law on short scales and delocalization for Wigner random matrices Semicircle law on short scales and delocalization for Wigner random matrices László Erdős University of Munich Weizmann Institute, December 2007 Joint work with H.T. Yau (Harvard), B. Schlein (Munich)

More information

NONLOCAL DIFFUSION EQUATIONS

NONLOCAL DIFFUSION EQUATIONS NONLOCAL DIFFUSION EQUATIONS JULIO D. ROSSI (ALICANTE, SPAIN AND BUENOS AIRES, ARGENTINA) jrossi@dm.uba.ar http://mate.dm.uba.ar/ jrossi 2011 Non-local diffusion. The function J. Let J : R N R, nonnegative,

More information

A path integral approach to the Langevin equation

A path integral approach to the Langevin equation A path integral approach to the Langevin equation - Ashok Das Reference: A path integral approach to the Langevin equation, A. Das, S. Panda and J. R. L. Santos, arxiv:1411.0256 (to be published in Int.

More information

Gaussian Process Approximations of Stochastic Differential Equations

Gaussian Process Approximations of Stochastic Differential Equations Gaussian Process Approximations of Stochastic Differential Equations Cédric Archambeau Dan Cawford Manfred Opper John Shawe-Taylor May, 2006 1 Introduction Some of the most complex models routinely run

More information

Short-time expansions for close-to-the-money options under a Lévy jump model with stochastic volatility

Short-time expansions for close-to-the-money options under a Lévy jump model with stochastic volatility Short-time expansions for close-to-the-money options under a Lévy jump model with stochastic volatility José Enrique Figueroa-López 1 1 Department of Statistics Purdue University Statistics, Jump Processes,

More information

Smoothers: Types and Benchmarks

Smoothers: Types and Benchmarks Smoothers: Types and Benchmarks Patrick N. Raanes Oxford University, NERSC 8th International EnKF Workshop May 27, 2013 Chris Farmer, Irene Moroz Laurent Bertino NERSC Geir Evensen Abstract Talk builds

More information

Kernel Sequential Monte Carlo

Kernel Sequential Monte Carlo Kernel Sequential Monte Carlo Ingmar Schuster (Paris Dauphine) Heiko Strathmann (University College London) Brooks Paige (Oxford) Dino Sejdinovic (Oxford) * equal contribution April 25, 2016 1 / 37 Section

More information

Lecture 2: From Linear Regression to Kalman Filter and Beyond

Lecture 2: From Linear Regression to Kalman Filter and Beyond Lecture 2: From Linear Regression to Kalman Filter and Beyond Department of Biomedical Engineering and Computational Science Aalto University January 26, 2012 Contents 1 Batch and Recursive Estimation

More information

Two recent works on molecular systems out of equilibrium

Two recent works on molecular systems out of equilibrium Two recent works on molecular systems out of equilibrium Frédéric Legoll ENPC and INRIA joint work with M. Dobson, T. Lelièvre, G. Stoltz (ENPC and INRIA), A. Iacobucci and S. Olla (Dauphine). CECAM workshop:

More information

Advanced computational methods X Selected Topics: SGD

Advanced computational methods X Selected Topics: SGD Advanced computational methods X071521-Selected Topics: SGD. In this lecture, we look at the stochastic gradient descent (SGD) method 1 An illustrating example The MNIST is a simple dataset of variety

More information

Asymptotic stability of an evolutionary nonlinear Boltzmann-type equation

Asymptotic stability of an evolutionary nonlinear Boltzmann-type equation Acta Polytechnica Hungarica Vol. 14, No. 5, 217 Asymptotic stability of an evolutionary nonlinear Boltzmann-type equation Roksana Brodnicka, Henryk Gacki Institute of Mathematics, University of Silesia

More information

ON COMPOUND POISSON POPULATION MODELS

ON COMPOUND POISSON POPULATION MODELS ON COMPOUND POISSON POPULATION MODELS Martin Möhle, University of Tübingen (joint work with Thierry Huillet, Université de Cergy-Pontoise) Workshop on Probability, Population Genetics and Evolution Centre

More information

Monte Carlo Methods. Handbook of. University ofqueensland. Thomas Taimre. Zdravko I. Botev. Dirk P. Kroese. Universite de Montreal

Monte Carlo Methods. Handbook of. University ofqueensland. Thomas Taimre. Zdravko I. Botev. Dirk P. Kroese. Universite de Montreal Handbook of Monte Carlo Methods Dirk P. Kroese University ofqueensland Thomas Taimre University ofqueensland Zdravko I. Botev Universite de Montreal A JOHN WILEY & SONS, INC., PUBLICATION Preface Acknowledgments

More information

A Backward Particle Interpretation of Feynman-Kac Formulae

A Backward Particle Interpretation of Feynman-Kac Formulae A Backward Particle Interpretation of Feynman-Kac Formulae Pierre Del Moral, Arnaud Doucet, Sumeetpal Singh To cite this version: Pierre Del Moral, Arnaud Doucet, Sumeetpal Singh. A Backward Particle Interpretation

More information

Perturbed Proximal Gradient Algorithm

Perturbed Proximal Gradient Algorithm Perturbed Proximal Gradient Algorithm Gersende FORT LTCI, CNRS, Telecom ParisTech Université Paris-Saclay, 75013, Paris, France Large-scale inverse problems and optimization Applications to image processing

More information

Extremal process associated with 2D discrete Gaussian Free Field

Extremal process associated with 2D discrete Gaussian Free Field Extremal process associated with 2D discrete Gaussian Free Field Marek Biskup (UCLA) Based on joint work with O. Louidor Plan Prelude about random fields blame Eviatar! DGFF: definitions, level sets, maximum

More information

Entropy-dissipation methods I: Fokker-Planck equations

Entropy-dissipation methods I: Fokker-Planck equations 1 Entropy-dissipation methods I: Fokker-Planck equations Ansgar Jüngel Vienna University of Technology, Austria www.jungel.at.vu Introduction Boltzmann equation Fokker-Planck equations Degenerate parabolic

More information

Machine Learning for Data Science (CS4786) Lecture 24

Machine Learning for Data Science (CS4786) Lecture 24 Machine Learning for Data Science (CS4786) Lecture 24 Graphical Models: Approximate Inference Course Webpage : http://www.cs.cornell.edu/courses/cs4786/2016sp/ BELIEF PROPAGATION OR MESSAGE PASSING Each

More information

Computer Intensive Methods in Mathematical Statistics

Computer Intensive Methods in Mathematical Statistics Computer Intensive Methods in Mathematical Statistics Department of mathematics johawes@kth.se Lecture 16 Advanced topics in computational statistics 18 May 2017 Computer Intensive Methods (1) Plan of

More information

Hypocoercivity for kinetic equations with linear relaxation terms

Hypocoercivity for kinetic equations with linear relaxation terms Hypocoercivity for kinetic equations with linear relaxation terms Jean Dolbeault dolbeaul@ceremade.dauphine.fr CEREMADE CNRS & Université Paris-Dauphine http://www.ceremade.dauphine.fr/ dolbeaul (A JOINT

More information

The Hierarchical Particle Filter

The Hierarchical Particle Filter and Arnaud Doucet http://go.warwick.ac.uk/amjohansen/talks MCMSki V Lenzerheide 7th January 2016 Context & Outline Filtering in State-Space Models: SIR Particle Filters [GSS93] Block-Sampling Particle

More information

Monte Carlo Approximation of Monte Carlo Filters

Monte Carlo Approximation of Monte Carlo Filters Monte Carlo Approximation of Monte Carlo Filters Adam M. Johansen et al. Collaborators Include: Arnaud Doucet, Axel Finke, Anthony Lee, Nick Whiteley 7th January 2014 Context & Outline Filtering in State-Space

More information

Nested Uncertain Differential Equations and Its Application to Multi-factor Term Structure Model

Nested Uncertain Differential Equations and Its Application to Multi-factor Term Structure Model Nested Uncertain Differential Equations and Its Application to Multi-factor Term Structure Model Xiaowei Chen International Business School, Nankai University, Tianjin 371, China School of Finance, Nankai

More information