### Table 1: Data partitions on difierent clusters. # of computing nodes # of rows # of columns

2003

### Table 1. Rectangular test matrices. Matrix Rows Columns Nonzeros

1998

"... In PAGE 7: ...he symmetrized matrix as was done by Berry et al. [3]. 4 Experimental Results In this section we compare the various partitioning methods presented in Sect. 3 on a collection of matrices listed in Table1 . These matrices were obtained from Matrix Market 2 with the exception of ccealink, man1, man2, and nhse400 which were provided by Michael Berry and are those used in Berry et al.... ..."

Cited by 11

### Table 3: Row, column, and

1994

Cited by 49

### Table 5: Comparison of the hybrid algorithm implementation and the row-column method, for P = 7. In Figure 5 we plot the computational time versus the size of the problem as well as the speedup ratio over the RC method applied to a data set zero padded up to the next power of two in each dimension. As 16

### Table 5 Sensitivity of the measures to row/column multiplications

1996

"... In PAGE 7: ... X2 and like- lihood ratio become bigger as the row totals or column totals become more balanced, while mutual information becomes smaller when the ratio of rst row/column becomes bigger. Table5 shows some examples, under the same grand total. In the case of bigram statistics, it seems intuitively more plausible to infer that the association of w1 and w2 is weaker for (a) than for (b), which in turn is weaker than for (c).... ..."

Cited by 1

### Table 1: The average (and stddev) percentage overlap in source IP addresses between (row, column) medium to large darknets over a month period.

2005

"... In PAGE 8: ... In order to see how the addition of darknets (each with their own number of unique source IP addresses over time) affects the aggregate number of sources in the hybrid mon- itor, we computed the overlap in unique source addresses between all darknets. Table1 shows the average percent- age of daily source IP address overlap (and standard devi- ation) in several of the medium to large darknets from the IMS system over a period of a month. A significant num- ber of the source IP addresses seen at these darknets are not globally visible.... In PAGE 9: ... on the detection time of remote events is a well-studied phenomena [22]. Table1 does show larger blocks with overlapping source IP addresses. However, the largest of these, although highly variable, only sees an average of 50% overlap.... ..."

Cited by 7

### Table 1: The average (and stddev) percentage overlap in source IP addresses between (row, column) medium to large darknets over a month period.

2005

"... In PAGE 8: ... In order to see how the addition of darknets (each with their own number of unique source IP addresses over time) affects the aggregate number of sources in the hybrid mon- itor, we computed the overlap in unique source addresses between all darknets. Table1 shows the average percent- age of daily source IP address overlap (and standard devi- ation) in several of the medium to large darknets from the IMS system over a period of a month. A significant num- ber of the source IP addresses seen at these darknets are not globally visible.... In PAGE 9: ... on the detection time of remote events is a well-studied phenomena [22]. Table1 does show larger blocks with overlapping source IP addresses. However, the largest of these, although highly variable, only sees an average of 50% overlap.... ..."

Cited by 7

### Table 6 Sensitivity of the measures to rst constant row/column ratio

1996

"... In PAGE 7: ... Thirdly, X2, Yule apos;s Y and mutual information monotonously decrease when the num- bers of observations in the rst row and the rst column increase while the grand total and the proportion of the three cells (in the rst row and in the rst column) are kept constant. Thus both in (a1) and (a2) and in (b1) and (b2) in Table6 , the values for these three measures decrease in the latter cases ((a2) and (b2)). On the other hand, the behaviour of likelihood ratio depends on the total size of the data, or more precisely, the estimated probability of the event.... ..."

Cited by 1

### Table 2: Coloring numbers (co.) and row/column lengths

### Table 5. Minumum and Maximum Hamming Distance information between rows and columns for 15-bit BCH code and 19-bit hybrid code with 105 codewords and classification accuracy for the Industry Sector dataset.

2000

"... In PAGE 7: ... The theoretical accuracy of the ECOC method can then be calculated by using the binomial distribution. A sample calculation for the first row of Table5 would be: Since Hmin=5, Emax= (5-1)/2=2 , even if 2 of the 15 classifiers give the wrong classification, the correct classification can still be obtained. The probability of at most 2 classifiers classifying incorrectly is: P(0 classifiers classify incorrectly) + P(1 classifier classifies incorrectly) + P(2 classifiers classify incorrectly)= 2 13 1 14 15 ) 846 .... ..."

Cited by 22