Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Cosmological parameter estimation using Particle Swarm Optimization (PSO) Jayanti Prasad in colloboration with Tarun Souradeep Inter-University Centre for Astronomy & Astrophysics (IUCAA) Pune, India (411007) August 11, 2011 Jayanti Prasad (IUCAA-PUNE) Parameter estimating using PSO August 11, 2011 1 / 14 CMBR data analysis pipeline Jayanti Prasad (IUCAA-PUNE) Parameter estimating using PSO August 11, 2011 2 / 14 “In theory at least, individual members of the school can profit from the discoveries and previous experience of all other members of the school during the search for food. This advantage can become decisive, outweighing the disadvantages of competition for food items, whenever the resource is unpedictably distributed in patches” [in reference to fish schooling] - E. O. Wilson, 1975, Sociobiology: The new synthesis Jayanti Prasad (IUCAA-PUNE) Parameter estimating using PSO August 11, 2011 3 / 14 Searching for the global maximum Jayanti Prasad (IUCAA-PUNE) Parameter estimating using PSO August 11, 2011 4 / 14 Parameter estimation Search for the maximum likelihood point (ML). Grid based methods - computational cost - exponentially Stochastic methods - computational cost - linearly (MCMC) Artificial Intelligence (AI) based optimization methods - Artificial Neural Network [Phillips & Kogut (2001) ], Genetic Algorithm [ Yao & Sethares (1994)], Particle Swarm Optimization [ Kennedy & Eberhart (1995)] We demonstrate the application of PSO for parameter estimation from WMAP 7 years data. Jayanti Prasad (IUCAA-PUNE) Parameter estimating using PSO August 11, 2011 5 / 14 Particle Swarm Optimization (PSO) Particle Swarm Optimization (PSO) Equation of Motion : i i Xt+1 = Xti + Vt+1 (1) i i Vt+1 = wVti + c1 ξ1 (Xti − XPbest ) + c2 ξ2 (Xti − XGbest ) (2) Velocity : w: c1 , c2 : ξ1 , ξ2 : XPbest : XGbest : inertia weight acceleration coefficients random numbers location of Pbest location of Gbest [Kennedy & Eberhart (1995); Wang & Mohanty (2010)] Jayanti Prasad (IUCAA-PUNE) Parameter estimating using PSO August 11, 2011 6 / 14 Particle Swarm Optimization (PSO) Implementation Initial conditions: i i i X i (t = 1) = Xmin + ξ(Xmax − Xmin ) i V i (t = 1) = ξVmax (3) where i i i Vmax = cv (Xmax − Xmin ) (4) Boundary conditions (reflecting boundary): Vti = −Vti ( i i Xmax , if Xti > Xmax and Xti = i i Xmin , if Xti < Xmin (5) − F(XGbest ,t )| < for t = 1, nstop (6) Stopping criteria: |F(XGbest Jayanti Prasad (IUCAA-PUNE) ,t+1 Parameter estimating using PSO August 11, 2011 7 / 14 Results Sampling Jayanti Prasad (IUCAA-PUNE) Parameter estimating using PSO August 11, 2011 8 / 14 Results Gbest Jayanti Prasad (IUCAA-PUNE) Parameter estimating using PSO August 11, 2011 9 / 14 Results Gbest Jayanti Prasad (IUCAA-PUNE) Parameter estimating using PSO August 11, 2011 10 / 14 Results Power spectrum Jayanti Prasad (IUCAA-PUNE) Parameter estimating using PSO August 11, 2011 11 / 14 Results Parameters Jayanti Prasad (IUCAA-PUNE) Parameter estimating using PSO August 11, 2011 12 / 14 summary Summary & conclusions Parameter estimation is an important exercise in current cosmological research. We have demonstrated that PSO also can be used for parameter estimation from CMBR data. Parameters which we have estimated using PSO are consistent with that obtained from other commonly used methods like MCMC. Due to simple algorithm and few design parameters, PSO can be easily implemented and parallelized on a cluster system for accelerated search of parameters. PSO can be very useful in situations when the dimensionality of the search space is very high and/or there are a large number of local maxima present. Jayanti Prasad (IUCAA-PUNE) Parameter estimating using PSO August 11, 2011 13 / 14 summary Thank You ! Jayanti Prasad (IUCAA-PUNE) Parameter estimating using PSO August 11, 2011 14 / 14 summary References Kennedy, J., & Eberhart, R. 1995, IEEE, 1942 Phillips, N. G., & Kogut, A. 2001, ArXiv Astrophysics e-prints Wang, Y., & Mohanty, S. D. 2010, Phys. Rev. D , 81, 063002 Yao, L., & Sethares, W. A. 1994, IEEE Transactions on Signal Processing, 927 Jayanti Prasad (IUCAA-PUNE) Parameter estimating using PSO August 11, 2011 14 / 14