### Table 5: Comparison of Margins from Local Optimization with Margins from RS

"... In PAGE 9: ...9 volume shown in Table5 is still below the global optimum, reflecting some very small constraint violations. Table 4.... In PAGE 9: ...5 140300 The areas and volumes of the single-level and two-level designs are shown in Table 4 above. Table5 shows the design variables for each beam and their margins from RSA and the lower-level optimization. We can see that margins from the lower-level optimization are off by about 2% for the first two beams, but there is a 10% violation for beam No.... In PAGE 9: ... 2 and No. 3 were increased by about one percent (see Table5 ), and the process of component optimization was repeated. The results are shown in Table 6.... ..."

### Table 1: Comparison with Soft Margin and Maximal Margin Parzen

2000

"... In PAGE 5: ... We also used bold driv- ing (Bishop 1995) to speed up the convergence. Table1 lists the errors for various settings of and C. For the synthetic data the error is the average true Bayes error over ten data sets, where the optimal parameter was chosen for each data set separately.... ..."

Cited by 13

### Table 1: Comparison with Soft Margin and Maximal Margin Parzen

2000

"... In PAGE 5: ... We also used bold driv- ing (Bishop 1995) to speed up the convergence. Table1 lists the errors for various settings of and C. For the synthetic data the error is the average true Bayes error over ten data sets, where the optimal parameter was chosen for each data set separately.... ..."

Cited by 13

### Table 1: Data for P using the Optimal Stability Margin Controller

2000

"... In PAGE 11: ...4.2). The results appear in Table 2. The stability margin corresponding to the plant and controller combination is necessarily inferior to the optimal stability margin bopt (compare the fourth column of the Table 2 to the sixth column of Table1 ) and so the frequency independent quantities in the fth and sixth columns of Table 2, which display (P( i); P( ))b-1 [p;C] and (P( i); P( ))b-1 [p;C] are necessarily worse than the corresponding data in Table 1. However, the frequency dependent quantities with the controller (4.... In PAGE 11: ...4.2). The results appear in Table 2. The stability margin corresponding to the plant and controller combination is necessarily inferior to the optimal stability margin bopt (compare the fourth column of the Table 2 to the sixth column of Table 1) and so the frequency independent quantities in the fth and sixth columns of Table 2, which display (P( i); P( ))b-1 [p;C] and (P( i); P( ))b-1 [p;C] are necessarily worse than the corresponding data in Table1 . However, the frequency dependent quantities with the controller (4.... ..."

Cited by 9

### Table 1: Data for P using the Optimal Stability Margin Controller

2000

"... In PAGE 9: ...2) Using Vinnicombe metric techniques, we verify that this was a good solution. Table1 shows the results of these calculations. The second column shows the i corresponding to each plant in the model set, and the next column shows a range of which corresponds to plants deemed to be \close to quot; i (these were calculated as = (1:2)i-0:5 and = (1:2)i+0:5).... In PAGE 9: ...e now investigate the plant-controller controller combination suggested by equation (4.2). The results appear in Table 2. The stability margin corresponding to the plant and controller combination is necessarily inferior to the optimal stability margin bopt (compare the fourth column of the Table 2 to the sixth column of Table1 ) and so the frequency independent quantities in the fth and sixth columns of Table 2, which display (P( i); P( ))b-1 [p;C] and (P( i); P( ))b-1 [p;C] are necessarily worse than the corresponding data in Table 1. However, the frequency dependent quantities with the controller (4.... In PAGE 9: ...e now investigate the plant-controller controller combination suggested by equation (4.2). The results appear in Table 2. The stability margin corresponding to the plant and controller combination is necessarily inferior to the optimal stability margin bopt (compare the fourth column of the Table 2 to the sixth column of Table 1) and so the frequency independent quantities in the fth and sixth columns of Table 2, which display (P( i); P( ))b-1 [p;C] and (P( i); P( ))b-1 [p;C] are necessarily worse than the corresponding data in Table1 . However, the frequency dependent quantities with the controller (4.... ..."

Cited by 9

### Table 1: Comparison with Soft Margin and Maximal Margin Parzen Pima

2000

"... In PAGE 5: ... We also used bold driv- ing (Bishop 1995) to speed up the convergence. Table1 lists the errors for various settings of a109 and a35 . For the synthetic data the error is the average true Bayes error over ten data sets, where the optimal parameter was chosen for each data set separately.... ..."

Cited by 13

### Table 3. Optimal classifiers for the customer insolvency problem

2006

"... In PAGE 19: ... Study of total gain for the classifiers resulting from the Neural Network algorithm Cost of LLR Classifier -2000 -1500 -1000 -500 0 500 0 5 10 15 20 25 Relative Gain CP/CN T G/C N LLR_5 LLR_10 LLR_25 LLR_50 LLR_1 Figure 20. Study of total gain or the classifiers resulting from the Linear Logistic Regression algorithm Following this procedure for all classifiers examined in this section we end up with the set exhibited in Figure 22 (with the solid line) and also in Table3 . These classifiers are the optimal ones, each for some specific interval of values of the fraction / P N CC, shown in the 1st column of Table 3.... In PAGE 19: ... Study of total gain or the classifiers resulting from the Linear Logistic Regression algorithm Following this procedure for all classifiers examined in this section we end up with the set exhibited in Figure 22 (with the solid line) and also in Table 3. These classifiers are the optimal ones, each for some specific interval of values of the fraction / P N CC, shown in the 1st column of Table3 . It appears that the classifiers that resulted from the voting schemes introduced in section 6 dominate the classifiers that generated them with the exception of NN_25 that exhibits superior performance in the interval (8.... In PAGE 20: ...igure 21. Breakeven points for all classifiers. Optimal Classifiers -100 100 300 500 700 900 1100 0 5 10 15 20 25 30 35 40 C P/CN TG / CN Optimal R2_50 R2_25 R1_25 NN_25 R3_15 R3_10 R3_5 R1_1 R3_1 R1_1 R2_50 R2_25 R1_25 NN_25 R3_15 R3_10 R3_5 Figure 22. The optimal classifiers for the customer insolvency problem In addition, in Table3 the performance of each classifier in the optimal list is indicated with respect to the performance measures examined in section 5. We can see that for small values of / P N CC (between 2.... In PAGE 22: ...Table 3. Optimal classifiers for the customer insolvency problem Lastly, from Table3 , it is important to understand the relationship between the class distribution in the training set and the optimal classifiers. When the gain from the correctly classified positives is a small multiple of the cost from the incorrectly classified negatives (e.... ..."

Cited by 2

### Table 5: The optimal correspondency on classifiers and features

1997

"... In PAGE 15: ...n decision rules in Eq.(7), Eq.(23), Eq.(29), Eq.(32) and Eq.(34) were also selected to find the best identifying result by trial and test. As a result, We found an optimal correspondency from 16 possible correspondencies on the current speech database and the optimal correspondency is listed in Table5 . Using the optimal correspon- dency, we found that all combination methods described in the paper could yield the best identifying results.... ..."

Cited by 20

### Table 5: The optimal correspondency on classifiers and features

1997

"... In PAGE 15: ...als defined in decision rules in Eq.(7), Eq.(23), Eq.(29), Eq.(32) and Eq.(34) were also selected to find the best identifying result by trial and test. As a result, We found an optimal correspondency from 16 possible correspondencies on the current speech database and the optimal correspondency is listed in Table5 . Using the optimal correspondency, we found that all combination methods described in the paper could yield the best identifying results.... ..."

Cited by 20