### Table 1: Bridge description: Speci cation and derived properties

1995

"... In PAGE 7: ... The candidate designs generated by Ecobweb generally do not satisfy important geometric constraints of the speci cation, such as length and clearances, and must be adapted. This adaptation uses a second hierarchy, called the derived hierarchy, created by Ecobweb from arti cial examples composed of the speci cation properties and the derived properties shown in Table1 . For example, one derived property is: ST-RATIO = TOWER?H SPAN?M .... In PAGE 8: ...properties, su cient for preliminary design purposes, are generated. Table1 provides their descriptions and acronyms. Redesign.... In PAGE 11: ...2 The analysis and redesign tasks The analysis results of the four candidates are given in Table 5. In the Table, the 12 design performance properties, introduced in Table1 , are expressed on a qualitative scale, where the values: UNDER, SERVICE, LIMIT, and EXCESSIVE denote: below 60%, 60-90%, 90-100%, and above 100% of the speci ed performance limit, respectively. In the Figure, the at segments of the in uence lines to the left of the tower in candidate 1 indicate that this span has a support under each stay connection (SIDE-S = 1).... In PAGE 14: ...peci cation. A value 1 of the Scaling measure indicates a perfect match. The combination of retrieval and adaptation is tested by measuring the Quality of the candidate designs after their adaptation, but before any redesign took place. The Quality measure is the summation of the deviation of 12 design performance properties shown in Table1 from their speci ed limits, penalizing the properties that exceed their respective limits more severely than those which Journal of Structural Engineering, ASCE, in press, July 1995 issue... ..."

Cited by 6

### Table 1: Sample sparse term-document matrix speci cations.

"... In PAGE 9: ... Depending upon the size of the database from which the term-document is generated, the matrix A can have several thousand rows and slightly fewer columns. Table1 lists a few statistics of ten sample sparse term-document matrices that have been generated2. We note that r and c are the average number of nonzeros per row and column, respectively.... In PAGE 9: ... We note that r and c are the average number of nonzeros per row and column, respectively. The Density of each sparse matrix listed in Table1 is de ned to be the ratio (Rows Columns) = (Nonzeros). By using the reduced model in (2), usually with k = n ( :01), minor di erences in terminology are virtually ignored.... In PAGE 10: ... As discussed in [3] and [8], LSI using the sparse SVD can be more robust and economical than straight term overlap methods. However, in practice, one must compute at least 100-200 largest singular values and corresponding singular vectors of sparse matrices having similar characteristics to those matrices in Table1 . In addition, it is not necessarily the case that rank(A) = n for the m n term-document matrix A, this is due to errors caused by term extraction, spelling, or duplication of documents.... In PAGE 10: ... In addition, it is not necessarily the case that rank(A) = n for the m n term-document matrix A, this is due to errors caused by term extraction, spelling, or duplication of documents. Regarding the numerical precision of the desired singular triplets for LSI, recent tests using a few of the databases listed in Table1 have revealed that the i-th residual, ~ ri, corresponding to the i-th approximate singular triplet, f~ ui; ~ i; ~ vig, need only satisfy 10?6 k~... In PAGE 33: ... (V0; RT 0 = 0 initially) For i = 2; 3 : : : ; k do: (k = bc=bc) (1a) Compute Yi?1 = ATAVi?1 ? Vi?1Si?1 ? Vi?1RT i?2, (1b) Orthogonalize Yi?1 against fVlgi?1 l=0, (1c) Factor Yi?1 = ViRi?1, (1d) Compute Si = V T i ATAVi. Table1 0: Hybrid Lanczos outer iteration used in bls2. where = diag f 1; 2; : : : ; ng, and i is an approximation to an exact singular value of the original m n sparse matrix A.... In PAGE 36: ...33 code opb() opm() opa() opat() bls1 { { y = Ax y = ATx bls2 y = AT Ax Y = ATAX y = Ax { Table1 1: Matrix-vector multiplication kernels for bls1 and bls2. 4 SVDPACKC Interface Before presenting our SVDPACKC benchmarks in Section 5, we illustrate how a simple yet e ective interface allows users to easily generate a se- ries of experiments using any or all of the SVDPACKC codes.... In PAGE 37: ... 5.1 Sparse Matrix Test Suite The 29 matrices listed in Table1 2, which arise from information retrieval and linear programming applications, were obtained from Apple Computer Inc., Cupertino.... In PAGE 37: ... The 16 remaining sparse rectangular matrices were extracted from a set of linear programming test problems compiled at Stanford Uni- versity [23]. From Table1 2, we can see that all of these matrices are less than 1% dense. We note that r and c are the average number of nonzeros per row and column, respectively.... In PAGE 41: ...4 2.8 Table1 2: SVDPACKC Sparse Matrix Test Suite. IR Information Re-... In PAGE 42: ...37 3.6 Table1 3: Machine Speci cations for SVDPACKC Benchmarks. 5.... In PAGE 42: ... We also provide tabulated results for both the Macintosh II/fx and Sun-4/490 in Tables 14 through 18 in Appendix A (Section 8) along with the number of approximated singular triplets, p, having residual norms (4) no larger than 10?6. Figures 4 through 7 (and Tables 14 through 17) re ect timings using the IR matrices from Table1 2, while Figure 8 (and Table 18) show elapsed user CPU times... In PAGE 43: ... However, las2 is still about 8 and 5 times faster than sis2 across both machines consid- ered.For the 16 LP matrices, we nd (see Figure 8 and Table1 8) the most competitive methods from the Lanczos-based group flas1, las2, bls1, bls2g and subspace iteration-based group fsis1, sis2, tms1, tms2g to be las2 and sis2, respectively. On both the Macintosh II/fx and Sun-4/490, las2 is on average 5 times faster than sis2 when computing as many as 50 of the largest singular triplets for the LP matrices arising from linear programming applications.... In PAGE 53: ...9 242.8 Table1 4: User CPU time (in seconds) expired by the single-vector Lanc- zos methods (las1, las2) on the Sun-4/490 and the Macintosh II/fx when... In PAGE 54: ...9 2037.2 Table1 5: User CPU time (in seconds) expired by the block Lanczos methods (bls1, bls2) on the Sun-4/490 and the Macintosh II/fx when computing the... In PAGE 56: ...1 2598.6 Table1 7: User CPU time (in seconds) expired by the trace minimization methods (tms1, tms2) on the Sun-4/490 and the Macintosh II/fx when com-... In PAGE 57: ...2 157.8 Table1 8: User CPU time (in seconds) expired by sis2 and las2 on the Sun- 4/490 and the Macintosh II/fx when computing the p-largest singular triplets... ..."

### Table 1: Sample sparse term-document matrix speci cations.

"... In PAGE 9: ... Depending upon the size of the database from which the term-document is generated, the matrix A can have several thousand rows and slightly fewer columns. Table1 lists a few statistics of ten sample sparse term-document matrices that have been generated2. We note that r and c are the average number of nonzeros per row and column, respectively.... In PAGE 9: ... We note that r and c are the average number of nonzeros per row and column, respectively. The Density of each sparse matrix listed in Table1 is de ned to be the ratio (Rows Columns) = (Nonzeros). By using the reduced model in (2), usually with k = n ( :01), minor di erences in terminology are virtually ignored.... In PAGE 10: ... As discussed in [3] and [8], LSI using the sparse SVD can be more robust and economical than straight term overlap methods. However, in practice, one must compute at least 100-200 largest singular values and corresponding singular vectors of sparse matrices having similar characteristics to those matrices in Table1 . In addition, it is not necessarily the case that rank(A) = n for the m n term-document matrix A, this is due to errors caused by term extraction, spelling, or duplication of documents.... In PAGE 10: ... In addition, it is not necessarily the case that rank(A) = n for the m n term-document matrix A, this is due to errors caused by term extraction, spelling, or duplication of documents. Regarding the numerical precision of the desired singular triplets for LSI, recent tests using a few of the databases listed in Table1 have revealed that the i-th residual, ~ ri, corresponding to the i-th approximate singular triplet, f~ ui; ~ i; ~ vig, need only satisfy 10?6 k~... In PAGE 33: ... (V0; RT 0 = 0 initially) For i = 2; 3 : : : ; k do: (k = bc=bc) (1a) Compute Yi?1 = ATAVi?1 ? Vi?1Si?1 ? Vi?1RT i?2, (1b) Orthogonalize Yi?1 against fVlgi?1 l=0, (1c) Factor Yi?1 = ViRi?1, (1d) Compute Si = V T i ATAVi. Table1 0: Hybrid Lanczos outer iteration used in bls2. where = diag f 1; 2; : : : ; ng, and i is an approximation to an exact singular value of the original m n sparse matrix A.... In PAGE 36: ...33 code opb() opm() opa() opat() bls1 { { y = Ax y = ATx bls2 y = AT Ax Y = ATAX y = Ax { Table1 1: Matrix-vector multiplication kernels for bls1 and bls2. 4 SVDPACKC Interface Before presenting our SVDPACKC benchmarks in Section 5, we illustrate how a simple yet e ective interface allows users to easily generate a se- ries of experiments using any or all of the SVDPACKC codes.... In PAGE 37: ... 5.1 Sparse Matrix Test Suite The 29 matrices listed in Table1 2, which arise from information retrieval and linear programming applications, were obtained from Apple Computer Inc., Cupertino.... In PAGE 37: ... The 16 remaining sparse rectangular matrices were extracted from a set of linear programming test problems compiled at Stanford Uni- versity [23]. From Table1 2, we can see that all of these matrices are less than 1% dense. We note that r and c are the average number of nonzeros per row and column, respectively.... In PAGE 41: ...4 2.8 Table1 2: SVDPACKC Sparse Matrix Test Suite. IR Information Re-... In PAGE 42: ...37 3.6 Table1 3: Machine Speci cations for SVDPACKC Benchmarks. 5.... In PAGE 42: ... We also provide tabulated results for both the Macintosh II/fx and Sun-4/490 in Tables 14 through 18 in Appendix A (Section 8) along with the number of approximated singular triplets, p, having residual norms (4) no larger than 10?6. Figures 4 through 7 (and Tables 14 through 17) re ect timings using the IR matrices from Table1 2, while Figure 8 (and Table 18) show elapsed user CPU times... In PAGE 43: ... However, las2 is still about 8 and 5 times faster than sis2 across both machines consid- ered.For the 16 LP matrices, we nd (see Figure 8 and Table1 8) the most competitive methods from the Lanczos-based group flas1, las2, bls1, bls2g and subspace iteration-based group fsis1, sis2, tms1, tms2g to be las2 and sis2, respectively. On both the Macintosh II/fx and Sun-4/490, las2 is on average 5 times faster than sis2 when computing as many as 50 of the largest singular triplets for the LP matrices arising from linear programming applications.... In PAGE 53: ...9 242.8 Table1 4: User CPU time (in seconds) expired by the single-vector Lanc- zos methods (las1, las2) on the Sun-4/490 and the Macintosh II/fx when... In PAGE 54: ...9 2037.2 Table1 5: User CPU time (in seconds) expired by the block Lanczos methods (bls1, bls2) on the Sun-4/490 and the Macintosh II/fx when computing the... In PAGE 56: ...1 2598.6 Table1 7: User CPU time (in seconds) expired by the trace minimization methods (tms1, tms2) on the Sun-4/490 and the Macintosh II/fx when com-... In PAGE 57: ...2 157.8 Table1 8: User CPU time (in seconds) expired by sis2 and las2 on the Sun- 4/490 and the Macintosh II/fx when computing the p-largest singular triplets... ..."

### Table 1. Size of the Speci cation

"... In PAGE 1: ... It also modularizes the speci cation, per- mitting reuse. The approximate sizes (without including comments and blank lines) of the major parts in the current TLA+ speci - cation are shown in Table1 . We do not plan to model primitives whose behavior depends on the underlying operating system, The framework plus our TLA+ models can be downloaded from [3].... ..."

### Table 2: Hardware speci cation.

"... In PAGE 4: ... Due to the duration of production runs, we do not account for ap- plication, mesh, or optimization con guration time, which we expect to be amortized over many time steps. We collected the results while running on the dedicated POWER3 node described in Table2 [25]. In addition to the architectural features listed above, the POWER3 implements a hardware- based prefetch engine that detects sequential instruction and data accesses and prefetches up to four streams simultane- ously.... ..."

### Table 1. Resource speci cations.

"... In PAGE 5: ... We have created an experiment based on the EU DataGRID Testbed 1, as shown in Figure 6 [9]. Table1 summarizes the characteristics of simulated re- sources, which were obtained from a real LCG testbed [15]. The parameters regarding to a CPU rating is de ned in the form of MIPS (Million Instructions Per Second) as per SPEC (Standard Performance Evaluation Corporation) benchmark.... In PAGE 5: ... Finally, each resource node has four CPUs. For this experiment, we have ve VO domains and each resource belongs to one of them as shown in Table1 . The... ..."

### Table 1: Example of cooperative activitytype speci cation

1995

"... In PAGE 5: ...ircuit. The example scenario is illustrated in Figure 3. A speci cation language can be developed based on the VODAK Model Language [KAN94] or the language in [RB94] or [KS94]. In the example demonstrated in Table1 , we use an intuitive pseudo notation to illustrate parts of the speci cation of cooperative activitytypes. An execution scenario of this example is presented in section 3.... ..."

Cited by 38

### Table 1. Knowledge-level speci cation of ACL primitives Agent primitives Knowledge-level behaviour

"... In PAGE 9: ... The agents has no constraints on the implementation language or knowledge representation formalisms it adopts, but it reacts to a well de ned protocol based on the standard primitives of an agent com- munication language. The primitives of this ACL [5] can be divided into four categories as shown in Table1 . Contents based services requests are realized as one-to-many primitives: whenever an agent needs a given service which solve a task T it can execute these multicast primitives.... ..."

### Table 1: Machine speci cations and latencies assumed.

"... In PAGE 5: ... 5 Performance Evaluation The underlying architectural model is a base su- perscalar processor which allows OOO execution, dy- namic speculation and performs writeback in in-order fashion. The detailed speci cations of the machine pa- rameters and latencies are speci ed in Table1 . The register les comprises of 64 registers each 64-bits wide, out-of-which only 32 registers are visible to the external user.... ..."

### Table 1: System Speci cations for the Target Architecture

1994

"... In PAGE 5: ... Some invalidations will be generated for absent sublines because processor caches may have had the subline but already evicted it due to collision or capacity misses. Table1 summarizes relevant system parameters for our experiments, with access times reported in the absence of contention. We simulated small application sizes in order to obtain reasonable simulation times and adjusted the cache sizes for each application based on the relevant application working set sizes.... ..."

Cited by 2