This paper describes Pairwise Bisection a nonparametric approach to optimizing a noisy function with few function evaluations The algorithm uses nonparametric reasoning about simple geometric relationships to nd minima e ciently Two factors often frus trate optimization noise and cost Output can contain signi cant quantities of noise or error while time or money allows for only a handful of experiments Pairwise bisection is used here to attempt to automate the pro cess of robust and e cient experiment design Real world functions also tend to violate tra ditional assumptions of continuousness and Gaussian noise Since nonparametric statis tics do not depend on these assumptions this algorithm can optimize a wide variety of phenomena with fewer restrictions placed on noise The algorithm s performance is com pared to that of three competing algorithms Amoeba PMAX and Q on several di erent test functions Results on these functions in dicate competitive performance and superior resistance to noise
[1]
M. Kendall.
A NEW MEASURE OF RANK CORRELATION
,
1938
.
[2]
Harold J. Kushner,et al.
wchastic. approximation methods for constrained and unconstrained systems
,
1978
.
[3]
George E. P. Box,et al.
Empirical Model‐Building and Response Surfaces
,
1988
.
[4]
Daniel N. Hill,et al.
An Empirical Investigation of Brute Force to choose Features, Smoothers and Function Approximators
,
1992
.
[5]
Leslie Pack Kaelbling,et al.
Learning in embedded systems
,
1993
.
[6]
Andrew W. Moore,et al.
Memory-based Stochastic Optimization
,
1995,
NIPS.
[7]
Marcos Salganicoff,et al.
Active Exploration and Learning in real-Valued Spaces using Multi-Armed Bandit Allocation Indices
,
1995,
ICML.
[8]
William H. Press,et al.
Numerical recipes in C
,
2002
.