1. Stochastic Approximation and Simulated Annealing Lecture 8 Leonidas Sakalauskas Institute of Mathematics and Informatics Vilnius, Lithuania EURO Working Group on Continuous Optimization

2.  Introduction.  Stochastic Approximation:  SPSA with Lipschitz perturbation operator;  SPSA with Uniform perturbation operator;  Standard Finite Difference Approximation algorithm.  Simulated Annealing  Implementation and Applications  Wrap-Up and Conclusions Content

3. In many practical problems of technical design some of the data may be subject to significant uncertainty which is reduced to probabilistic- statistical models. The performance of such problems can be viewed like constrained stochastic optimization programming tasks. Stochastic Approximation can be considered as alternative to traditional optimization methods, especially when objective functions are no differentiable or computed with noise. Introduction

4. Application of Stochastic Approximation to solving of optimization problems, while the objective function is non-differentiable or nonsmooth and computed with noise is a topical theoretical and practical problem. The known methods of Stochastic Approximation for solving of these problems use the idea of stochastic gradient and certain rules of changing of step length for ensuring the convergence. Stochastic Approximation

5. The optimization problem is (minimization) as follows: where is a bounded from below Lipshitz function. Formulation of the optimization problem

6. Let be generalized gradient of this function. Assume to be a set of stationary points: and to be a set of function values: Formulation of the optimization problem

7. We consider a function smoothed by perturbation operator: where is the value of the perturbation parameter. The functions smoothed by this operator are twice continuously differentiable (Rubinstein & Shapiro (1993), Bartkute & Sakalauskas (2004)), that offers certain opportunities creating optimization algorithms.

8. At last time the interesting research was focussed on Stochastic Perturbation Stochastic Approximation (SPSA) It is enough to calculate values of the function only in one or some points for the estimation of the stochastic gradient in SPSA algorithms, that promises for us to reduce numerical complexity of optimization. Advantages of SPSA

9. 1. SPSA with Lipschitz perturbation operator. 2. SPSA with Uniform perturbation operator. 3. Standard Finite Difference Approximation algorithm. SA algorithms

10. General Stochastic Approximation scheme where stochastic gradient and This scheme is the same for different Stochastic Approximation algorithms whose distinguish only by approach for stochastic gradient estimation.

11. SPSA with Lipschitz perturbation operator Gradient estimator of the SPSA with Lipschitz perturbation operator is expressed as: where - is the value of the perturbation parameter, -is uniformly distributed in the unit ball vector -is the volume of the n-dimensional ball ( Bartkute & Sakalauskas ( 2007 ))

12. SPSA with Uniform perturbation operator Gradient estimator of the SPSA with Uniform perturbation operator is expressed as: where -is the value of the perturbation parameter, -is a vector consisting of variables uniformly distributed from the interval [-1;1] (Mikhalevitch et al (1987)).

13. Standard Finite Difference Approximation algorithm Gradient estimator of the Standard Finite Difference Approximation algorithm is expressed as: where -is the value of the perturbation parameter, -is uniformly distributed in the unit ball; vector -is the vector with zero components except i th one, which is equal to 1. (Mikhalevitch et al (1987)).

14. Let consider that the function f(x) has a sharp minimum in the point, in which the algorithm converges when Then where A>0, H>0, K>0 are certain constants, is minimum point of the smoothed function. Rate of convergence

15. The proposed methods were tested with following functions: where is a set of real numbers randomly and uniformly generated in the interval, Computer simulation The samples of T=500 test functions were generated, when

16. Empirical and theoretical rates of convergence by SA methods Theoretical rates 1.51.751.9 Empirical rates SPSA (Lipshitz perturbation) n = 2 1.455091.720131.892668 n = 4 1.418011.744261.958998 SPSA ( Uniform perturbation) n = 2 1.6052441.9383191.988265 n = 4 1.5514861.7845191.998132 Stochastic Difference Approximation method n = 2 1.527991.763991.90479 n = 4 1.502361.750571.90621

17. The rate of convergence (n = 2)

18. The rate of convergence (n = 10)

19. Let us consider the application of SA to the minimization of the mean absolute pricing error for the parameter calibration in the Heston Stochastic Volatility model [Heston S. L.(1993)]. We consider the mean absolute pricing error (MAE) defined as : Volatility estimation by Stochastic Approximation algorithm where N is the total number of options, and the realized market price and the implied the theoretical model price, respectively, while (n=6) are the parameters of the Heston model to be estimated. represent

20. To compute option prices by the Heston model, one needs input parameters that can hardly be found from the market data. We need to estimate the above parameters by an appropriate calibration procedure. The estimates of the Heston model parameters are obtained by minimizing MAE: Let consider the Heston model for the Call option on SPX (29 May 2002).

21. Minimization of the mean absolute pricing error by SPSA and SFDA methods

22. In cargo oil tankers design, it is necessary to choose such sizes for bulkheads, that the weight of bulkheads would be minimal. Optimal Design of Cargo Oil Tankers

23. subject to The minimization of weight of bulkheads for the cargo oil tank we can formulate like nonlinear programing task (Reklaitis et al (1986)): where - width,-debt, - lenght, - thikness.

24. SPSA with Lipschitz perturbation for the cargo oil target design

25. Confidence bounds of the minimum (A=6.84241, T=100, N=1000)

26. Simulated Annealing Global optimization methods  Global algorithms (bounds and branch algorithms, dynamic programming, full selection, etc)  Greedy optimization (local search)  Heuristic optimization

27. Metaheuristics  Simulated Annealing  Genetic Algorithms  Swarm Intelligence  Ant Colony  Taboo search  Scatter search  Variable neighborhood  Neural Networks  Etc.

28. Simulated Annealing algorithm Simulated Annealing algorithm is developed by modeling steel annealing process (Metropolis et al. (1953)) A lot of applications in Operational Research and Data Analysis, etc.

29. Simulated Annealing Main idea: to simulate drift of current solution with probability distribution to improve solution updating - temperature function - neighborhood function

30. Simulated Annealing algorithm Step 1. Choose,,, set. Step 2. Generate drift with probability distribution Step 3. If and (Metropolis rule) then accept: ; otherwise Step 2

31. Improvement of SA by Pareto Type models The theoretical investigation of SA convergence shows, that in these algorithms Pareto type models can be applied to form search sequence (Yang (2000)). Class of Pareto models, main feature and parameter: Pareto model’s distributions have "heavy tails“. α - the main parameter of these models, which impacts the heaviness of the tail α –stable distributions are Pareto (follows to C.L.T.)

32. Pareto type (Heavy-tailed) distributions Main features: infinite variance, infinite mean Introduced by Pareto in the 1920’s Mandelbrot established the use of heavy-tailed distributions to model real-world fractal phenomena. There are a lot of other applications (financial market, traffic in computer and telecommunication networks, etc.).

33. Pareto type (Heavy-tailed) distributions Decay of Distributions Heavy-Tailed - Power Law (polynomial) Decay (e.g. Pareto-Levy): where 0 0 are constants

34. - stable distributions

35. Comparison of tail probabilities for standard normal, Cauchy and Levy distributions In this table were compared the tail probabilities for the three distributions. It is clear that the tail probability for the normal quickly becomes negligible, whereas the other two distributions have a significant probability mass in the tail.

36. Improvement of SAS by Pareto type models The convergence conditions (Yang (2000)) indicate that, under suitable conditions, an appropriate choice of the temperature and neighborhood size updating functions ensures the convergence of the SA algorithm to the global minimum of the objective function over the domain of interest. The following corollaries give different forms of temperature and neighborhood size updating functions corresponding to different kinds of generation probability density functions to guarantee the global convergence of the SA algorithm.

37. Convergence of Simulated Annealing

38. Improvement of SA in continuous optimization The above corollaries indicate that a different form of temperature updating function has to be used with respect to a different kind of generation probability density function in order to ensure the global convergence of the corresponding SA algorithm.

39. Convergence of Simulated Annealing Some Pareto-type models were explored. See the Table 1.

40. Convergence of Simulated Annealing

41. Testing of SA for continuous optimization In global and combinatorial optimization problems, when optimization algorithms are used, the reliability and efficiency of these algorithms is needed to be tested. Special testing functions, known in literature, are used for this. Some of these functions have one or more global minimum, some of them have global and local minimums. With the help of these functions it can be ensured, that the methods are efficient enough, thus, it is possible to test and prevent algorithms from being trapped in local minimum, as well as the speed and accuracy of convergence and other parameters can be watched.

42. Testing criteria By modeling SA algorithm with some testing functions with two different distributions, and changing some optional parameters, there were some questions: which of these distributions guarantees the faster convergence to global minimum by value of objective function; what are probabilities of finding global minimum, how can impact these probabilities the changing of some parameters; what the proper number of iterations, which guarantees the finding global minimum with desirable probability.

43. Testing criteria value of minimized objective function; probability to find global minimum after some number of iterations. These characteristics were computed by Monte-Carlo method - N realizations (N=100, 500, 1000) with K iterations each (K=100, 500, 1000, 3000, 10000, 30000). Characteristics evaluated by Monte-Carlo simulation:

44. Testing functions An example of test function: Branin’s RCOS (RC) function (2 variables): RC(x1,x2)=(x2-(5/(42))x12+(5/)x1- 6)2+10(1-(1/(8)))cos(x1)+10; Search domain: 5 < x1 < 10, 0 < x2 < 15; 3 minima: (x1, x2)*=(-, 12.275), (, 2.275), (9.42478, 2.475); RC((x1, x2)*)=0.397887.

45. Simulation resulsts

46. Simulation results

48. Fig. 1. Probability to find global minimum by SA for Rastrigin function

49. 1. The SA methods have been considered for comparison SPSA with Lipschitz perturbation operator; SPSA with Uniform perturbation operator and SFDA method as well Simulated Annealing; 2. Computer simulation by Monte-Carlo method has shown that the empirical estimates of the rate of convergence of SA for nondifferentiable functions corroborate the theoretical rates Wrap-Up and Conclusions