"... In PAGE 8: ... Drill-down Queries. Our implementation considers four types of queries (shown in Table4 ) that involve different extents of spatio- temporal processing that evaluate both advantages and limitations of wavelet compression. The GlobalYearlyEdge and LocalYear- lyMean queries explore features for which wavelet processing is typically well suited.... In PAGE 9: ... Our goal in this section is to demonstrate the search features of the system and prove our claim that multi-resolution storage can be useful for a broad variety of queries. To evaluate performance, each of the queries shown in Table4 was posed over the dataset. For yearly queries (GlobalYearlyEdge and and GlobalYearlyMax), there were 45 instances each, since there are 45 years of data.... In PAGE 10: ... Figure 6.2 shows the variation of query quality for queries defined in Table4 for different levels of drill-down. Performance for LocalYearlyMean, GlobalYearlyMax and Local- DailyMax queries are very similar, as shown in Figure 6.... In PAGE 10: ... Evaluating Training using Limited Information In our evaluation, we use a training period of two epochs of data (10% of total deployment time) to predict the query accuracy for the entire dataset. Summaries are constructed over the training set, and all queries in Table4 are posed over these summaries. Ideally, the error obtained from the training set would mirror error seen by the omniscient scheme.... In PAGE 11: ... The first column in Table 7 shows the difference between the per- formance of training and the optimal schemes. These results are ag- gregate results over a range of storage sizes (0 - 100KB) and query types (shown in Table4 ). Training performs exceedingly well, and in fact is on average less than 1% worse than the optimal solution.... ..."

"... In PAGE 8: ... Drill-down Queries. Our implementation considers four types of queries (shown in Table4 ) that involve different extents of spatio- temporal processing that evaluate both advantages and limitations of wavelet compression. The GlobalYearlyEdge and LocalYear- lyMean queries explore features for which wavelet processing is typically well suited.... In PAGE 9: ... Our goal in this section is to demonstrate the search features of the system and prove our claim that multi-resolution storage can be useful for a broad variety of queries. To evaluate performance, each of the queries shown in Table4 was posed over the dataset. For yearly queries (GlobalYearlyEdge and and GlobalYearlyMax), there were 45 instances each, since there are 45 years of data.... In PAGE 9: ... Figure 6.2 shows the variation of query quality for queries defined in Table4 for different levels of drill-down. Performance for LocalYearlyMean, GlobalYearlyMax and Local- DailyMax queries are very similar, as shown in Figure 6.... In PAGE 10: ... Evaluating Training using Limited Information In our evaluation, we use a training period of two epochs of data (10% of total deployment time) to predict the query accuracy for the entire dataset. Summaries are constructed over the training set, and all queries in Table4 are posed over these summaries. Ideally, the error obtained from the training set would mirror error seen by the omniscient scheme.... In PAGE 11: ... The first column in Table 7 shows the difference between the per- formance of training and the optimal schemes. These results are ag- gregate results over a range of storage sizes (0 - 100KB) and query types (shown in Table4 ). Training performs exceedingly well, and in fact is on average less than 1% worse than the optimal solution.... ..."

"... In PAGE 12: ... In addition, various types of particle interactions can be distinguished. As shown in Table1 , A supports ve stable particles, and one unstable \particle quot;, , which occurs at SS boundaries1. \lives quot; for only one time step, after which it decays into two other particles, and , respectively occurring at SD and DS boundaries.... ..."

"... In PAGE 15: ... The reactions between particles and depend on the relative distance d between them, where 0 d 2r. The most striking feature to be observed in Table2 is its similarity to the domain, particle, and particle interaction table for rule A. Indeed the number of domains, the number of particles, and the number and types of particle interactions are identical in the two tables.... ..."

"... In PAGE 8: ... 4.2 Query Performance Can good query accuracy be obtained by processing summaries constructed with low communication overhead? We address this question by evaluating the performance of drill-down processing on a broad set of queries as shown in Table5 . Each of... In PAGE 9: ... Figure 4.2 shows the variation of query quality for queries defined in Table5 for different levels of drill-down. Both the LocalYearlyMean query and the two Max queries are processed as regular drill-down queries.... In PAGE 10: ... Such data might be expected to be available if a data gathering phase preceded actual deployment. Summaries are constructed over the training set, and queries in Table5 are posed over these summaries. Ideally, the error obtained from the training set would mirror error seen by the omniscient scheme.... ..."