### Table 8 Attained signi cance levels (in %) in using the original Cox statistic (Cox), the Victoria-Feser statistic (VF) and the Marazzi statistic (M) in using a xed amount of contamination (Pareto against Exponential).

### Table 15 Attained signi cance levels (in %) in using the Cox statistic (Cox), the Victoria-Feser statistic (VF) and the Marazzi statistic (M) with a xed amount of contamination (Gamma against Weibull).

### Table 7 Comparison of the contamination using a xed amount of outliers and the -contamination, where represents the amount of outliers. In our simulations we xed this amount and in using the -contamination one gets in average per cent of outliers. The di er- ence listed in this table represents the average di erence (over 1000 samples) between outliers.

### Table 1: Amount of state transferred for SSES. The total includes xed overheads.

"... In PAGE 5: ... This includes pixel information as well as side information such as macroblock mode. Table1 gives the amount of data required for the the state transfer. The ma- jor components are the YUV pixel information of the refer- ence frames for the original video, required for performing reconstruction at the new transcoder; macroblock informa- tion for the reference frames; as well as YUV pixel informa- tion for the reference frames for the down-sampled video.... ..."

### Table 3: Graphs selected for measurement. We vary surface-to-volume ratio by changing the dimension of the graph and examine both disconnected and highly connected graphs. Each graph uses a xed amount of memory per processor (memory constrained scaling [29]). The last column shows the percentage of nodes in the largest component.

1995

Cited by 11

### Table 1: Functional Unit Kind and Latency it useful to introduce the notion of \cut-o convergence quot; which constrains the maximum number of iterations scheduled to some xed amount; the remainder of any paths that have not converged after the cut-o number of iterations are simply scheduled sequentially. We stress that \cut-o convergence quot; is a creature of our experiment|its purpose is to identify when code explosion is a problem. In practice one should prefer to use scheduling heuristics designed to prevent code explosion; this topic is discussed further in Section 7.2.

1991

"... In PAGE 32: ... With the exception of two \unlimited resource quot; experiments used to measure threshold performance results, the register le is assumed to have 32 registers. Operation latencies for both models, given in Table1 , are similar to the Motorola 88110 Superscalar. An instance of the heterogeneous model has 1, 2, 3, or an unlimited number of each of the functional units de ned in Table 1.... In PAGE 32: ... Operation latencies for both models, given in Table 1, are similar to the Motorola 88110 Superscalar. An instance of the heterogeneous model has 1, 2, 3, or an unlimited number of each of the functional units de ned in Table1 . An instance of the homogeneous model has 2, 4, 8, or an unlimited number of homogeneous functional units, where each homogeneous functional unit can perform any of the functions de ned in Table 1.... ..."

Cited by 29

### Table 12 represents the attained signi cance levels in considering several amounts of contamination and several values of the nominal level. One remarks that the Cox statistic always rejects H0 for contaminated samples. This result is not surprising as the breakdown point, i.e. the highest percentage of outliers that will not completely distort the estimate, of the maximum likelihood estimate ^ equals 0, and as with an amount of contamination of 3% we already have 6 outliers in our sample of size 200. Moreover, one notices that even the Victoria-Feser statistic, which uses the maximum likelihood estimate as well, over rejects H0 at = 3%. This is not the case with the Marazzi statistic which remains stable at amounts of outliers up to 5%.

"... In PAGE 22: ... Table12 Attained signi cance levels (in %) in using the original Cox statistic (Cox), the Victoria-Feser statistic (VF) and the Marazzi statistic (M) in considering a xed amount of contamination (Expo- nential against Pareto). Table 13 represents the attained powers.... ..."

### Table 13 Attained power in using the original Cox statistic (Cox), the Victoria-Feser statistic (VF) and the Marazzi statistic (M) with xed contamination (Exponential against Pareto).

"... In PAGE 22: ...4 Table 12 Attained signi cance levels (in %) in using the original Cox statistic (Cox), the Victoria-Feser statistic (VF) and the Marazzi statistic (M) in considering a xed amount of contamination (Expo- nential against Pareto). Table13 represents the attained powers. One observes that, once again, the Marazzi statistic is the most powerful.... ..."

### Table 6. Comparison of xed-base window and xed-base comb methods for Fp192. w is the window width, S denotes the number of points stored in the precomputation phase, and T denotes the number of eld operations.

2001

"... In PAGE 15: ... Return(Q). From Table6 we see that the xed-base comb method is expected to slightly outperform the xed-base window method for similar amounts of storage. For our implementation, we chose w =4forthecombmethodandw =5for xed- base window for curves over Fp192, Fp224,andFp256; the curves over the larger elds Fp384 and Fp521 used w =5forcombandw = 6 in xed-base window.... ..."

Cited by 33