"... In PAGE 34: ... 6.3 Results on real-world instances In Table6 we give the results obtained on instances derived from the real- world transportation problem. In this case too we can note a difference in the performance of the two algorithms.... ..."

"... In PAGE 30: ... 6.2 Results on randomly created instances In Table3 we present the results obtained by running our algorithms on in- stances derived from the CVRP literature. The rst columns give the name of the original instance, the number of customers, the class and the total number of items, respectively.... ..."

"... In PAGE 8: ...worth noting that there is only one minimally redundant NST algorithm for every radix b. Table1 summarizes the values of , and for these special algorithms, for power of two radices: 2 b 32. Notice that the only possible radix 2 NST division ( = 1, = 0, = 1 2) is Burgess apos; algorithm [7], where rj 1rj 2 are recoded in the following two cases: 11 = 0 1 and 1 1 = 01.... In PAGE 8: ... However, no matter what the analysis is, the quotient-digit selection function comes up to be the same as that of the general development presented in this paper. An algorithm not listed in Table1 and deserving special attention concerns the digit-set D lt;16:10 gt;, because all the digits in such a digit-set can be represented as the sum of two digits from the radix 4 signed-digit-set D lt;4:2 gt; (i.... ..."

"... In PAGE 4: ... As an important feature, the developing trend of the wireless network architecture is base on an all-IP core with asymmetry accesses, where mobile ad hoc networks can be treated as a special class of access networks since many mobile ad hoc networks have some internetworking with other infrastructures like the Internet. From the cellular structure point of view, future mobile networks can be divided into different sizes of cellular coverage, as shown in Table1 . The basic idea behind this is to seamlessly integrate two categories of wireless network technologies together, i.... In PAGE 5: ... One access point usually covers a certain zone. 6) Nano-mobility, is the mobility within the zone covered by one access point, where the cell zone can be various from mega-cell down to nano-cell, see Table1 . One access point may employ several logical channels.... ..."

"... In PAGE 10: ... We remark that this scatter value is the only component of this algorithm that is based on a \distance quot; measure, and it would be just as easy to use other measures not based on a \distance quot; measure and appropriate for particular data sets. Having discussed all the components of our algorithm, we now summarize the overall algorithm in Table2 . At each pass through the main loop, we select a node based on our measure of \cohesiveness, quot; obtain the mean vector and principal direction for the documents associated with that node, and split the documents using the mean vector and principal direction into two children nodes.... In PAGE 11: ... In addition, typical values for snz in our examples range from :04 = 4% down to :0068 = :68%. The bulk of the cost in the algorithm of Table2 is the SVD computation in step 7. We have already mentioned that the main memory required for this step is space for ksvd m-vectors.... In PAGE 14: ... These are summarized in Table 3 [8]. We applied two algorithms to this data set, the divisive PDDP algorithm ( Table2 ), and an agglomerative algorithm [5]. The agglomerative algorithm is shown only for comparative purposes, and is brie y summarized as follows.... In PAGE 17: ...e joined. Further consolidation could occur using l = 2 or perhaps with larger values of k. This aspect deserves further investigation. 6 Extensions and Future Work The binary tree constructed by the PDDP algorithm of Table2 can be extended in several ways. In this section, we discuss several extensions that could be applied to this algorithm, including some preliminary experimental results validating some of them.... In PAGE 17: ... 6.1 Classi cation of New Documents The simplest extension to the PDDP algorithm of Table2 is to classify a new set of documents according to the clusters from an original document set by using the original binary PDDP... In PAGE 21: ... We selected the documents so that both document sets had representatives of every topic label. We used the method of Table2 on the rst half of the document set to construct a new PDDP tree with 16 leaf nodes, then we used the method of Table 8 to classify the remaining documents, so that we end up with a set of 16 clusters, each containing documents from both halves of the original document set. The resulting entropy of the combined result (i.... ..."