# A simulated annealing procedure based on the ABC Shadow algorithm for statistical inference of point processes

Recently a new algorithm for sampling posteriors of unnormalised probability densities, called ABC Shadow, was proposed in [8]. This talk introduces a global optimisation procedure based on the ABC Shadow simulation dynamics. First the general method is explained, and then results on simulated and real data are presented. The method is rather general, in the sense that it applies for probability densities that are continuously differentiable with respect to their parameters

## Authors

• 1 publication
• 1 publication
• 1 publication
• ### Shadow Simulated Annealing algorithm: a new tool for global optimisation and statistical inference

This paper develops a new global optimisation method that applies to a f...
07/15/2019 ∙ by R. Stoica, et al. ∙ 0

• ### Approximate Bayesian inference for a spatial point process model exhibiting regularity and random aggregation

In this paper, we propose a doubly stochastic spatial point process mode...
03/23/2020 ∙ by Ninna Vihrs, et al. ∙ 0

• ### An algorithm for computing Fréchet means on the sphere

For most optimisation methods an essential assumption is the vector spac...
02/26/2018 ∙ by Gabriele Eichfelder, et al. ∙ 0

• ### Some Remarks on Replicated Simulated Annealing

Recently authors have introduced the idea of training discrete weights n...
09/30/2020 ∙ by Vincent Gripon, et al. ∙ 0

• ### Fast nonparametric classification based on data depth

A new procedure, called DDa-procedure, is developed to solve the problem...
07/20/2012 ∙ by Tatjana Lange, et al. ∙ 0

• ### MESA: Maximum Entropy by Simulated Annealing

Probabilistic reasoning systems combine different probabilistic rules an...
03/13/2013 ∙ by Gerhard Paaß, et al. ∙ 0

• ### On a flexible construction of a negative binomial model

This work presents a construction of stationary Markov models with negat...
12/18/2018 ∙ by Fabrizio Leisen, et al. ∙ 0

##### This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

## 1 SA method : general description

The SA algorithm is a global optimisation method. Assume that the probability density is to be maximised. This is achieved by sampling while . If the temperature parameter goes to in an appropriate way, then the SA algorithm converges asymptotically towards the global optimum. This method is rather general. Under smooth assumptions, the algorithm can be generalised to minimise any criteria that can be written as .

SA algorithms for maximising probability densities for random fields and marked point process such as (1) are presented in [1, 4, 7]. The obtained cooling schedules for the temperature parameter are of the form

 T=T01+logn

with

. The solution guaranteed by the method converges towards the uniform distribution over the sub-space of configurations that maximises  (

1).

The difficulty of solving (3) is due to the fact that the normalising constant is not available in analytic closed form. Hence, special strategies are required to sample from the posterior distribution (2). The present paper use for thie purpose, the ABC Shadow simulation dynamics [8].

The ABC Shadow dynamics is an approximate algorithm able to sample posteriors. Its main steps are presented below :

Algorithm ABC Shadow : fix and . Assume the observed pattern is and the current state is .

1. Generate according to .

2. For to do

• Generate a new candidate following the density defined by

 UΔ(θ→ψ)=1VΔ1b(θ,Δ/2){ψ}, (4)

representing the uniform probability density over the ball of volume .

• The new state is accepted with probability given by

 αs(θ→ψ)= = min{1,p(ψ|y)p(θ|y)×f(x|θ)c(ψ)1b(ψ,Δ/2){θ}f(x|ψ)c(θ)1b(θ,Δ/2){ψ}},

otherwise .

3. Return

4. If another sample is needed, go to step with .

## 2 Results

The SA Shadow algorithm is applied here to the statistical analysis of patterns which are simulated from a Strauss model [3, 9]. This model describes random patterns made of points exhibiting repulsion. Its probability density is

 p(y|θ) ∝βn(y)γsr(y)= =exp[n(y)logβ+sr(y)logγ]. (5)

Here is a point pattern in the window , while and are the sufficient statistic and the model parameter vectors, respectively. The sufficient statistics components and represent respectively, the number of points in and the number of pairs of points at a distance closer than .

The Strauss model on the unit square and with density parameters , and , was considered. This gives for the parameter vector of the exponential model . The CFTP algorithm (see Chapter 11 in [6]) was used to get samples from the model and to compute the empirical means of the sufficient statistics . The SA based on the ABC Shadow algorithm was run using as observed data.

The prior density was the uniform distribution on the interval . Each time, the auxiliary variable was sampled using steps of a MH dynamics [5, 6]. The and parameters were set to and , respectively. The algorithm was run for iterations. The intial temperature was set to . For the cooling schedule a slow polynomial scheme was chosen

 Tn=kT⋅Tn−1

with . A similar scheme was chosen for the parameters, with . Samples were kept every steps. This gave a total of samples.

Figure 1 shows the results obtained after running the SA ABC Shadow based algorithm. The final values for and were and , respectively. These values are close to the true model parameters.

## 3 Conclusions and perspectives

The numerical results obtained are satisfactory. Actually, the algorithm is applied on real astronomical data, and the obtained models are tested and validated. Since the ABC Shadow is an approximate algorithm, the theoretical convergence of the SA procedure based on it, needs further investigation [2].

### Acknowledgments.

Part of the work of the first author was supported by a grant of the Romanian Ministry of National Education and Scientific Research, RDI Programme for Space Technology and Avanced Research - STAR, project number 513.

## References

• [1] Geman, S. and Geman, S (1984). Stochastic relaxation, Gibbs distributions and the Bayesian restoration of images. IEEE Transactions on Pattern Analysis and Machine Intelligence 6, 721–741.
• [2] Haario, H. and Saksman, E. (1991). Simulated annealing process in general state space. Advances in Applied Probability 23, 866–893.
• [3] Kelly, F. P. and Ripley, B. D. (1976). A note on Strauss’s model for clustering. Biometrika 63, 357–360.
• [4] van Lieshout, M. N. M. (1994). Stochastic annealing for nearest neighbour point processes with application to object recognition. Advances in Applied Probability 26, 281–300.
• [5] van Lieshout, M. N. M. (2000). Markov point processes and their Applications. Imperial College Press. London.
• [6] Møller, J. and WaagepetersenLamport, R. P. (2004). Statistical inference and simulation for spatial point processes. Chapman and Hall/CRC. Boca Raton.
• [7] Stoica, R. S., Gregori, P. and Mateu, J. (2005). Simulated annealing and object point processes : tools for analysis of spatial patterns. Stochastic Processes and their Applications 115, 1860–1882.
• [8] Stoica, R. S., Philippe, A., Gregori, P. and Mateu, J. (2017). ABC Shadow algorithm: a tool for statistical analysis of spatial patterns. Statistics and Computing 27, 1225–1238.
• [9] Strauss, D. J. (1975). A model for clustering. Biometrika 62, 467–475.