### Tables 2, present results that highlights the role played by one single feature of our algorithm. For each set of experiments the parameters associated with one specific feature are varied while all the other parameters are kept at the value of the best overall result. All the results are obtained making a large number of trials of shorter runs with GRSSHC. Specifically, we made 1000 trials for each configuration with MaxIters equal to 1000. What we measure is the average number of violations for the final state. The top-left part of Table 2 regards the look-ahead mechanism. We consider three possibilities for the computation of the look-ahead factor. Either no look-ahead factor is added to the cost function, or the number of available employees for a pair shift- task in which the requirements are not completely fulfilled are counted two different ways. The non-biased look-ahead counts cost 1 for each employee which cannot do the task, whereas in the biased version, the value is multiplied by the number of missing employees. Intuitively, the biased version emphasizes cases of more than one employee missing per task. The results show a mild advantage for using the biased version.

1999

Cited by 8

### Tables 2, present results that highlights the role played by one single feature of our algorithm. For each set of experiments the parameters associated with one specific feature are varied while all the other parameters are kept at the value of the best overall result. All the results are obtained making a large number of trials of shorter runs with GRSSHC. Specifically, we made 1000 trials for each configuration with MaxIters equal to 1000. What we measure is the average number of violations for the final state. The top-left part of Table 2 regards the look-ahead mechanism. We consider three possibilities for the computation of the look-ahead factor. Either no look-ahead factor is added to the cost function, or the number of available employees for a pair shift- task in which the requirements are not completely fulfilled are counted two different ways. The non-biased look-ahead counts cost 1 for each employee which cannot do the task, whereas in the biased version, the value is multiplied by the number of missing employees. Intuitively, the biased version emphasizes cases of more than one employee missing per task. The results show a mild advantage for using the biased version.

1999

Cited by 8

### Table 3. 5. The law of large numbers is neither a child nor the father of probability Many people have thought about infrequent (or nonexistent) communication between the A-world of probability and the R-world where the action is. This includes even those who, like Feller, have taken a fanatical position on \inde- pendence quot; (see the quote in the previous section). Feller writes [5, p. 141], as a preamble to motivating the \law of large numbers quot;, \On several occasions we have mentioned that our intuitive notion of [ab- stract] probability is based on the following assumption. If in n identical trials A occurs times, and if n is very large, then =n should be near the

"... In PAGE 14: ... Here the sets Z+ 2187 and fk (mod 9) 9 + Z+ 2187g are \almost quot; the same for all k. The digits are shown in Table3 . By displaying the digits in groups of ve Table 3 gives the appearance of a piece of a table of 105 random numbers (i.... In PAGE 14: ... The digits are shown in Table 3. By displaying the digits in groups of ve Table3 gives the appearance of a piece of a table of 105 random numbers (i.e.... ..."

### Table 2: Performance of Improve (averaged over 70 trials).

2000

"... In PAGE 14: ... This process is repeated until no suggested modi cation improves the plan. Table2 shows the performance of the IMPROVE algorithm, as reported in [16], on a large evacuation domain that contains 35 cities, 45 roads, and 100 people. The people are scattered randomly in each trial, and the goal is always to bring all the people, using two trucks and a helicopter, to one central location.... In PAGE 15: ... The HIGH approach replaces the PlanMine component of IMPROVE with a technique that simply tries to prevent the malfunctions that occur most often. As shown in Table2 , IMPROVE with PlanMine increases a plan apos;s probability of achieving its goal, on average, by about 15%, but without PlanMine only by, on average, about 3%. 5.... ..."

Cited by 2

### Table 2. Performance of Improve (averaged over 70 trials).

2000

"... In PAGE 15: ... This process is repeated until no suggested modi cation improves the plan. Table2 shows the performance of the IMPROVE algorithm, as reported in (Lesh, Martin, amp; Allen 1998), on a large evacuation domain that contains 35 cities, 45 roads, and 100 people. The people are scattered randomly in each trial, and the goal is always to bring all the people, using two trucks and a helicopter, to one central location.... In PAGE 16: ... The HIGH approach replaces the Plan- Mine component of IMPROVE with a technique that simply tries to prevent the malfunctions that occur most often. As shown in Table2 , IMPROVE with Plan- Mine increases a plan apos;s probability of achieving its goal, on average, by about 15%, but without PlanMine only by, on average, about 3%. 5.... ..."

Cited by 2

### Table 1 displays the estimate statistics for 250 trials obtained by ESPRIT, WSF, and the suboptimal SSF algorithm. The estimates for each trial were computed using 100 simulated snapshots from the array. Note that the variance of the azimuth and elevation estimates of the suboptimal approach are respectively 25% and 10% lower than those obtained by ESPRIT. WSF o ers a further reduction in variance of 25% for the elevation estimates, and a large improvement in the quality of the azimuth estimates.

"... In PAGE 20: ... Table1 : Estimate Statistics for Example 3... ..."

### Table 3. Validation of NCS on large cluster workload. Examines performance for HTTP/1.0 (with up to 4 parallel connections), HTTP/1.1 (with a single persistent connection), and HTTP/1.0+KeepAlive (with up to either 4 or 6 parallel connections for Netscape or MSIE respectively). Measured shows the mean of 5 trials, except for the last two rows where it is the mean of 3 trials, from [20].

2001

"... In PAGE 4: ...rror from 39.3% to 24.3%, although [15] was validating a slightly different dataset from an earlier version of the paper by Nielsen et al). Since [20] include many more measurements than those used by Heidemann et al, we provide additional compar- isons in Table3 . Again, the simulated values are not farfrom the measured values, except for the case of Ethernet (with likely similar reasons as those described earlier).... ..."

Cited by 6

### Table 3. Validation of NCS on large cluster workload. Examines performance for HTTP/1.0 (with up to 4 parallel connections), HTTP/1.1 (with a single persistent connection), and HTTP/1.0+KeepAlive (with up to either 4 or 6 parallel connections for Netscape or MSIE respectively). Measured shows the mean of 5 trials, except for the last two rows where it is the mean of 3 trials, from [20].

"... In PAGE 4: ...rror from 39.3% to 24.3%, although [15] was validating a slightly different dataset from an earlier version of the paper by Nielsen et al). Since [20] include many more measurements than those used by Heidemann et al, we provide additional compar- isons in Table3 . Again, the simulated values are not farfrom the measured values, except for the case of Ethernet (with likely similar reasons as those described earlier).... ..."

### Table 3. Validation of NCS on large cluster workload. Examines performance for HTTP/1.0 (with up to 4 parallel connections), HTTP/1.1 (with a single persistent connection), and HTTP/1.0+KeepAlive (with up to either 4 or 6 parallel connections for Netscape or MSIE respectively). Measured shows the mean of 5 trials, except for the last two rows where it is the mean of 3 trials, from [20].

"... In PAGE 4: ...rror from 39.3% to 24.3%, although [15] was validating a slightly different dataset from an earlier version of the paper by Nielsen et al). Since [20] include many more measurements than those used by Heidemann et al, we provide additional compar- isonsin Table3 . Again, the simulated valuesare not far from the measured values, except for the case of Ethernet (with likely similar reasons as those described earlier).... ..."

### Table 1: Percentage of correct model order selection (over 50 trials) by different criteria for synthetic Gaussian data with 75 and 700 samples respectively.

"... In PAGE 9: ... Examples of models selected by different criteria are shown in Figure 1(c). Table1 shows examples of the percentage of correct model order selection (over 50 trials) by different criteria give small and large sample sizes.... ..."