### Table 3 shows that, for a small target site sample size and for the quantiles Q0.75 and Q0.95, all 294

"... In PAGE 13: ...Table3... In PAGE 33: ...Table3 : Performance of BAY and IFL estimators for quantile Q0.... ..."

### Table 7. Simulation results for N=25 and low-noise block target function, when the sample size is very small model selection tasks are more difficult, in this case NIC shows a very high variance on the observed efficiency

### Table XIII presents the asymptotic performance bounds for the INTERF and POTENG sections on eight processors. For the INTERF section, the asymptotic performance of the dynamic feedback algorithm is very close to the performance of the optimal algorithm. For the POTENG section, however, the asymptotic performance difference is significant when the target production interval is small. Two factors contribute to this behavior. First, the mean minimum effective sampling interval for the Aggressive policy in the POTENG section is approximately 0:29 seconds, which is quite large compared to the target produc- tion interval. Second, the mean sampled overhead of the Aggressive policy is 0:91 for the POTENG section. This policy delivers almost no useful work during its sampling phases. The result is that for small target production intervals, the application spends a significant amount of its time executing a policy that delivers almost no useful work.

in Eliminating Synchronization Overhead in Automatically Parallelized Programs Using Dynamic Feedback

1999

Cited by 7

### Table XIII presents the asymptotic performance bounds for the INTERF and POTENG sections on eight processors. For the INTERF section, the asymptotic performance of the dynamic feedback algorithm is very close to the performance of the optimal algorithm. For the POTENG section, however, the asymptotic performance difference is significant when the target production interval is small. Two factors contribute to this behavior. First, the mean minimum effective sampling interval for the Aggressive policy in the POTENG section is approximately 0:29 seconds, which is quite large compared to the target produc- tion interval. Second, the mean sampled overhead of the Aggressive policy is 0:91 for the POTENG section. This policy delivers almost no useful work during its sampling phases. The result is that for small target production intervals, the application spends a significant amount of its time executing a policy that delivers almost no useful work.

in Eliminating Synchronization Overhead in Automatically Parallelized Programs Using Dynamic Feedback

1999

Cited by 7

### Table 3: Performance of BAY and IFL estimators for quantile Q0.75, Q0.95 and Q0.995. Target site sample size: 10.

"... In PAGE 13: ...1 BAY vs. IF L Approach 292 [ Table3 about here.] 293 Table 3 shows that, for a small target site sample size and for the quantiles Q0.... In PAGE 13: ... IF L Approach 292 [Table 3 about here.] 293 Table3 shows that, for a small target site sample size and for the quantiles Q0.75 and Q0.... ..."

### Table 2 shows the target densities and the average solid ink density of 20 samples at each inking level. This table verifies if the press run conforms to the target densities. It was observed that discrepancies between the target density and the measured density are small, i.e., less than 0.05 with the exception of high inking levels (shown in Italized type). The density differences between the target and the high inking level were mainly caused by density dry back and the press inking control limits.

"... In PAGE 5: ... Table2 . Target densities and the average solid ink density of 20 samples at each inking level.... ..."

### Table 2: Only a small fraction of the hybridization sample even has an opportunity to find the appropriate target. This calculation assumes D #19 10 ,7 cm2 /s for ssDNA 1000 bp long and a micro-array with an area of 625 mm2 (1 inch by 1 inch). The efficiency of hybridization could be dramatically improved using a micro-fluidic transport chamber.

### Table 8: Assessment of the comparison sample of white drivers for a target sample of black drivers derived from the propensity weighting

2004

"... In PAGE 51: ... However, we were only able to find about 305 similarly situated white drivers so statistical power may be small. Table8 shows that after weighting, the black drivers group was still slightly more likely to be stopped in East Oakland, but the difference is not statistically large enough to cause concern. Table 8: Assessment of the comparison sample of white drivers for a target sample of black drivers derived from the propensity weighting ... ..."