### Table 1 gives the geometric mean of the cost ratios between two optimization algorithms, CGBoost and AnyBoost, at the final iteration. As we expected, the cost of CGBoost is generally much lower than that of AnyBoost.

2003

"... In PAGE 7: ... Table1 : The average final cost ratio of CGBoost to AnyBoost. Numbers less than BCBMBCBCBCBCBH are shown as BCBMBCBCBCBC.... ..."

Cited by 2

### Table 1 gives the geometric mean of the cost ratios between two optimization algorithms, CGBoost and AnyBoost, at the final iteration. As we expected, the cost of CGBoost is generally much lower than that of AnyBoost.

2003

"... In PAGE 7: ... Table1 : The average final cost ratio of CGBoost to AnyBoost. Numbers less than 0.... ..."

Cited by 2

### Table 2. Subproblem results of subinstance SUB4 with the algorithm SRDS limited to 18 variables. Each subproblem is build from the previous one including a new variable. The columns are: number of variables of the subproblem, time for optimally solving the subproblem in seconds, optimal cost of the subproblem (that is minardsia being i the variable included in this particular subproblem), maximum cost of all the values of the variable included (that is maxardsia), and the mean of the costs of the values of the new variable.

2001

"... In PAGE 12: ... For example, subinstance SUB4 is solved using SRDS up to a 18 variables sub- problem, for the following subproblems 19, 20, 21, and 22 variables, the limited version of SRDS is used instead. In Table2 the detailed execution for solving subinstance SUB4 is presented. We have observed that CELAR-6 subinstances are a perfect example of poor homogeneity within the specialized lower bounds for every value rdsia for the same variable i (in the same variable, values with high cost can coexist with consistent values with cost 0) and also it can happen that the mean value cost of a variable is high and the following (respect to the static order) is low; suggesting that the variable where to combine rds and ic that gives the best lower bound contribution is not necessary the previous variable in the static ordering.... ..."

Cited by 9

### Table 1: Cost standard deviation results Algorithm N % of costs gt; mean % of costs within one standard deviation EM 300 0:367 0:767

"... In PAGE 23: ... Analysis of the costs apos; standard deviations reveals two observations. First, Table1 tabulates the percentages for the two algorithms of the number of runs resulting in costs greater than the mean cost and the number of runs with costs within a standard deviation of the mean. The EM algorithm again performs slightly better in both accounts, but the results are comparable.... In PAGE 26: ... Examination of the errors apos; standard deviation yields additional observations. Table 2 contains information similar to that contained in Table1 . It lists the percentages for the two algorithms of the number of runs resulting in errors greater than the mean error and the number of runs with... ..."

### Table 2: Error standard deviation results Algorithm N % errors gt; mean % costs within one standard deviation EM 300 0:392 0:658

"... In PAGE 26: ... Examination of the errors apos; standard deviation yields additional observations. Table2 contains information similar to that contained in Table 1. It lists the percentages for the two algorithms of the number of runs resulting in errors greater than the mean error and the number of runs with... ..."

### Table 3. Performance Comparison Ave. Mean Cost/

### Table 4: Further tests on the QAP benchmark problems using the same perturbations and CPU times as before; given is the mean solution cost, averaged over BDBC independent runs for each instance. Here we consider three different choices for the acceptance criterion. Clearly, the inclusion of diversification significantly lowers the mean cost found.

### Table 3. No. of classes No. of runs Mean cost Std Dev

1997

"... In PAGE 6: ...The effect of initial modes on clustering results is shown in Table3 . The first column is the number of disease classes the initial modes have and the second is the corresponding number of runs with the number of disease classes in the initial modes.... ..."

Cited by 62