Universit´e Grenoble Alpes and ENSIMAG Master SIAM
Stochastic modelling for neurosciences
Lab 4: Particle filter
Objectives: Implementing a particle filter.
Let us consider a discretized auto-regressive model forj= 0, . . . , n, Xj+1=φXj+Uj
withUj ∼ N(0, τ2) and where the observation is
Vj =Xj+ηj
withηj ∼ N(0, σ2).
1. Simulaten= 100 observations withφ= 0.9,τ2= 0.1,σ2= 1 starting withX0=−10.
We want to filter X1:n from the observations V1:n.
2. Implement the particle filter using the transition density as proposal distribution, with K = 100,500,1000,2000 particles. Sample one filter trajectory and compare to the true simulatedX. 3. Implement the particle filter with the optimal proposal equal, at timej
N
τ2σ2 τ2+σ2
φXj−1
τ2 +Vj
σ2
, τ2σ2 τ2+σ2