Advanced Computer Graphics - Virginia：先进的计算机图形-弗吉尼亚.pptVIP

Advanced Computer Graphics Spring 2002 Professor Brogan Simulated Annealing Monte Carlo approach for minimizing multivariate functions Monte Carlo = Random = Stochastic Requires one ‘goodness’ metric – result of evaluation function Multivariate – selects multiple parameter values to minimize evaluation function Algorithm Outline Select some initial guess of evaluation function parameters: x0 Evaluate evaluation function, E(x0)=v Compute a random displacement, x’0 The Monte Carlo event Evaluate E(x’0) = v’ If v’ v; set new state, x1 = x’0 Else set x1 = x’0 with Prob(E,T) This is the Metropolis step Repeat with updated state and temp What is Annealing? Used to treat work-hardened parts made out of low-carbon steels Heat to a specific temperature, then soak, and then cool slowly Thermodynamics – molecules can move around when they are at high temps. Slow cooling permits self organization into minimum energy configurations Back to our Situation We approximate nature’s alignment of molecules by allowing uphill transitions exp (-E/kT) Boltzmann Probabilty Distribution Even when T is small, a disruption is possible exp (-(E2-E1) / kT) Metropolis Step The rate at which T is decreased and the amount it is decreased is prescribed by an annealing schedule What have we got? Always move downhill if possible Sometimes go uphill Optimality guaranteed with slow annealing schedule No need for smooth search space No derivatives Can be discrete search space Traveling salesman problem Optimization Given: What value of minimizes f()? Expensive to computer f()? Expensive to compute partials of f()=Jacobian? Global vs. local solutions Constrained Optimization Minimize Subject to: There are a priori limitations on the possible values of the independent variables Linear Programming Special type of constrained optimization Function f() and constraints are linear * *