### Table 1. Validating assumption: b gt; gt; t .

"... In PAGE 6: ... Data exchanged among p proces- sors in phase 3 (p=8), t=512 megabytes. Regarding the second assumption, Table1 shows t and bi, i = 2; 4; 8; 16, for a portion of the WSJ collection. The data confirm that the assumption is realistic.... ..."

### Table 1. Ontological assumptions of the critical realistic view of science (Bhaskar 1978). Xs indicate the domain of reality in which mechanisms, events, and experiences, respectively reside, as well as the domains involved for such a residence to be possible.

"... In PAGE 3: ... (Bhaskar 1989). Bhaskar (1978) outlines what he calls three domains: the real, the actual, and the empirical ( Table1 ). The real domain consists of underlying structures and mechanisms, and relations; events and behavior; and experiences.... ..."

### Table 3 Multigrid V -cycle performance. A Proof of regularity and approximation assumption In this section we prove the regularity and approximation assumption (35) for the rectangular RT0 mixed method from the computational results section. We note that the arguments can be extended to triangular grids, other mixed 19

2000

"... In PAGE 18: ... We solve the problem on a series of grid levels and domain decompositions to study the scalability of the multigrid algorithm. We run on three grid levels and ve di erent decompositions (see Table3 ). On the rst grid level the subdomain grids are taken to be 4 4 4 and 5 5 5 in a checkerboard fashion.... ..."

Cited by 3

### Table 1. Assumed realistic bandwidth distribution of MorphMix nodes and acceptable intermediate and final nodes.

"... In PAGE 8: ... Depending on the node level, we define acceptance probabilities, which is the probability a node accepts relaying further anonymous tunnels when it is contacted as a new neighbor by another node. The left half of Table1 illustrates the node levels and their up- and down-stream bandwidths, the distribution of MorphMix nodes over the node levels, and the acceptance probabilities. Note that these assumptions are only valid for honest nodes.... In PAGE 8: ...Table1... In PAGE 9: ...odes an acceptance probability of 0.8 and 0.95, respectively. Note that we have not explicitly listed nodes with Cable connections because the bandwidths they offer are the same as ADSL or DSL connections. Therefore, the ADSL and DSL nodes in Table1 also include nodes with Cable connections. A second valuable result from the measurement study are the up-times of the peers.... In PAGE 9: ... In practice, this means that the initiator specifies a minimum node level for the nodes it accepts and intermediate nodes offer only nodes in selections that meet or exceed this minimum level. The right half of Table1 specifies reasonable acceptable node levels for the intermediate and final nodes depending on the node level of the initiator. We analyze how well the collusion detection mechanism copes with the realis- tic acceptance and up-time probabilities defined above.... In PAGE 10: ... The table below the graphs give the opti- mal attack level (oal) and pam with and without collusion detection for malicious nodes in 1000, 2000, 5000, and 10000 subnets. We assume the initiator belongs to the four fastest types of nodes in Table1 , which corresponds to the worst case since the spectrum of nodes that can be offered in selections is smallest. The figures in parenthesis give pam if no tunnel optimization according to the right half of Table 1 were made, i.... In PAGE 10: ... We assume the initiator belongs to the four fastest types of nodes in Table 1, which corresponds to the worst case since the spectrum of nodes that can be offered in selections is smallest. The figures in parenthesis give pam if no tunnel optimization according to the right half of Table1 were made, i.e.... In PAGE 11: ... At first, this seems surprising because DSL512, T1, and T3 account for only 40% of all nodes, which means the effective number of honest nodes for fast initiators is much smaller because slow nodes are no longer offered in selections to them. But actually, not so much has changed because looking at the acceptance prob- abilities in Table1 shows that slow nodes accept relaying tunnels infrequently compared to the fast nodes, which means that even if they were accepted by fast initiators, they would be present in their tunnels rather infrequently. This implies that by requesting a minimum quality for the nodes offered in selection for fast nodes, we have merely removed occasional occurrences of slow nodes in these selections.... In PAGE 12: ... We analyze the download times for a complete web page depending on whether the web server is accessed directly or through MorphMix. In the latter case, we also compare the results with and without tunnel quality optimization according to Table1... In PAGE 13: ...ig. 2. Download times when accessing the web server directly and through MorphMix. The results are split into the six node levels defined in Table1 . Comparing Figures 2(a) and (b), we see that the download times get significantly longer if the web server is accessed through MorphMix without tunnel optimization.... ..."

### Table 1. Assumed realistic bandwidth distribution of MorphMix nodes and acceptable intermediate and nal nodes.

"... In PAGE 8: ... Depending on the node level, we de ne acceptance probabilities, which is the probability a node accepts relaying further anonymous tunnels when it is contacted as a new neighbor by another node. The left half of Table1 illustrates the node levels and their up- and down-stream bandwidths, the distribution of MorphMix nodes over the node levels, and the acceptance probabilities. Note that these assumptions are only valid for honest nodes.... In PAGE 8: ...Table1... In PAGE 9: ...odes an acceptance probability of 0.8 and 0.95, respectively. Note that we have not explicitly listed nodes with Cable connections because the bandwidths they o er are the same as ADSL or DSL connections. Therefore, the ADSL and DSL nodes in Table1 also include nodes with Cable connections. A second valuable result from the measurement study are the up-times of the peers.... In PAGE 9: ... In practice, this means that the initiator speci es a minimum node level for the nodes it accepts and intermediate nodes o er only nodes in selections that meet or exceed this minimum level. The right half of Table1 speci es reasonable acceptable node levels for the intermediate and nal nodes depending on the node level of the initiator. We analyze how well the collusion detection mechanism copes with the realis- tic acceptance and up-time probabilities de ned above.... In PAGE 10: ... The table below the graphs give the opti- mal attack level (oal) and pam with and without collusion detection for malicious nodes in 1000, 2000, 5000, and 10000 subnets. We assume the initiator belongs to the four fastest types of nodes in Table1 , which corresponds to the worst case since the spectrum of nodes that can be o ered in selections is smallest. The gures in parenthesis give pam if no tunnel optimization according to the right half of Table 1 were made, i.... In PAGE 10: ... We assume the initiator belongs to the four fastest types of nodes in Table 1, which corresponds to the worst case since the spectrum of nodes that can be o ered in selections is smallest. The gures in parenthesis give pam if no tunnel optimization according to the right half of Table1 were made, i.e.... In PAGE 11: ... At rst, this seems surprising because DSL512, T1, and T3 account for only 40% of all nodes, which means the e ective number of honest nodes for fast initiators is much smaller because slow nodes are no longer o ered in selections to them. But actually, not so much has changed because looking at the acceptance prob- abilities in Table1 shows that slow nodes accept relaying tunnels infrequently compared to the fast nodes, which means that even if they were accepted by fast initiators, they would be present in their tunnels rather infrequently. This implies that by requesting a minimum quality for the nodes o ered in selection for fast nodes, we have merely removed occasional occurrences of slow nodes in these selections.... In PAGE 12: ... We analyze the download times for a complete web page depending on whether the web server is accessed directly or through MorphMix. In the latter case, we also compare the results with and without tunnel quality optimization according to Table1... In PAGE 13: ...ig. 2. Download times when accessing the web server directly and through MorphMix. The results are split into the six node levels de ned in Table1 . Comparing Figures 2(a) and (b), we see that the download times get signi cantly longer if the web server is accessed through MorphMix without tunnel optimization.... ..."

### Table 4: Correlations between assumptions for stochastic optimization problem.

"... In PAGE 5: ...re independent of each other. This in reality, however, is rarely the case. These unit quantities, or coefficients, can and will exhibit variation, and the assumptions have some degree of relationship to each other. Consider Table 3, showing the modification of these coefficients reflecting the more realistic stochastic behavior, and Table4 , showing correlations between the assumptions. Determining a feasible combination of x1 and x2 so as to optimize the profit now becomes more difficult given that the coefficients exhibit variability, along with their interrelationships.... ..."

### Table 1: Summary of Square 2 results The other synthetic data is the more realistic Di- verging Tree image sequence. The underlying motion in this example is a bilinear expansion from the cen- ter of the image. We use the regularization parameter

"... In PAGE 5: ... The optical ow was obtained after 40 iterations of the incomplete Cholesky preconditioned conjugate gradi- ent algorithm. The errors of our modi ed gradient- based regularization method and the other gradient- based methods for this translating square example are summarized in Table1 . Only the results from those gradient-based methods that yield 100% density ow estimates reported in [2] are included in this table for comparison.... ..."

Cited by 1

### Table 1: Symbols used. (caches, servers, bandwidth etc.), i.e., there are no additional costs based on number of requests served. This is a reasonable assumption because servers incur xed costs and bandwidth can be bought at a at monthly rate. If mainte- nance costs are negligible or xed, pro t maximization is equivalent to revenue maximization. We also assume that the market is monopolistic, i.e., there is no other entity selling the same content. This is a realistic assumption in many scenarios where the content owner personally sells the content or has licensed it to a single distributor.

"... In PAGE 6: ... This is a realistic assumption in many scenarios where the content owner personally sells the content or has licensed it to a single distributor. Table1 presents the symbols we have used in our analysis. Consider an arbitrary customer who wants to purchase content pi;j.... ..."

Cited by 1

### Table 13: Two Functionally Distinct Language Processing (Sub)modules. The two (sub)modules together serve to distinguish regular from irregular verbs, and to predict the past tenses of unseen verbs, unmarked as regular or irregular. Generally speaking, both of these (sub)modules are similar in the sense that both regular and irregular verbs are computed by production rules. After all, irregular verbs can be regarded as regular with \rules quot; 41

in Answering the Connectionist Challenge: A Symbolic Model Of Learning the Past Tenses Of English Verbs

"... In PAGE 43: ... 1991 explanation of this e ect. Despite our doubts about how well our compression algorithms model the actual psycho- logical process of learning English past tense, the nal set of rules including the abstract rep- resentation of the verb stem and the verb endings using new predicates in Table13 seem to be a psychologically realistic. One problem with our model might be in accounting for the real-time implementation of blocking (Cf.... ..."