A note on estimating false alarm rates via importance sampling

Advances in Neyman-Pearson Neural Detectors Design

“…Suboptimal solutions for f * X (x) which partially minimize (14) are given in the literature, by scaling input signals [5], by shifting them [6,7], or by other strategies [8,9,10].…”

“…or any monotonic increasing function of (18); moreover, some researchers [9,10] propose fX(x|H 1 ) as IS density function. As in the decision region of…”

Efficient Simulation Of Random Signal Detectors

“…Orsak has shown that the gain associated with this importance sampling strategy is asymptotically exponential (i.e., produces a straight line in the log-log scale used in figure 2) [5]. However, there is no guarantee that the slope of this line will not be very small as it was for this example.…”

“…Many researchers have utilized importance sampling to reduce the number of trials in Monte Carlo simulations; some of the more recent results are contained in [4,5,8] which also reference many of the earlier works. However, the main hindrance to using many of these results has been the analytical complexity in determining a "good" importance sampling approach.…”

Applications of POD studies and robust design to electromagnetic NDE

“…The cost of importance sampling is that it is not easy to construct nonuniform random variables that have densities similar to p(y I x, ) and p(y l*o) respectively [116]. This problem has been solved in many applications in signal detection that estimate small probability of false alarm [117][118][119][120][121][122]. However, they are efficient only when the PFA is very small (generally smaller than l.e-4), which is not the case in current NDE techniques.…”