"... In PAGE 12: ...30 % 93.16 % D : external duration model, T : structural state-tying Table4 shows the increment of time complexity due to multiple models. Since search space of the recognition network broadened by the multiple model setting, average time for character recognition was ... ..."

"... In PAGE 15: ...ith tests of Schwarz preconditioning against the popular global ILU(k), k = 0; . . .5, family of preconditioners on a non-dedicated single-processor SUN SPARCstation with 512MB of memory. The results are summarized in Table5 . Because of the overlap and the coarse solve, the Schwarz preconditioner needs more memory, even if all subdomain problems are solved inexactly with ILU(5), than the other global ILU(k) preconditioners.... In PAGE 16: ...in this paper that is already in the local domain of convergence on the nest grid. (Observe, for instance, that the number of Newton steps required on M1 = 0:8 problem the 256 256 grid in Table5 , is roughly half that of the corresponding problem on the 512 512 grid in Table 3. If the shock is correctly located on a (relatively) coarse grid, the plateau of Fig.... ..."

"... In PAGE 18: ... We have found that the virtual machine emulation of typical image understanding applications takes from a few seconds to a few minutes and from 30 to 487 times faster than on the CAAPP instruction-level simulator. A sample of these results is shown in Table6 . The applications are described in [14].... ..."

"... In PAGE 18: ...Table1 shows the execution times of prototype implementations of their ap- proaches in seconds for three benchmark programs. The timings were taken on single processor SUN workstations.... ..."

"... In PAGE 18: ...Table1 shows the execution times of prototype implementations of their ap- proaches in seconds for three benchmark programs. The timings were taken on single processor SUN workstations.... ..."

"... In PAGE 9: ... Table2 shows the increment of complexity due to multiple models. The number of models was increased about 2.... In PAGE 9: ... As a result, the number of free parameters for the observation probability distributions becomes almost the same as that of the single model setting. Next, the third row of Table2 shows the increment of time complexity. Since search space of the recognition network broadened by the multiple model setting, average time for character recognition was increased about 2 times.... ..."

"... In PAGE 27: ...5 Table 3: Average iteration counts for Problem A with c gt;0 Therefore we restrict attention to d=2, and we assume that all distances between interpolation points are stored by the preliminary work, because they are required by every iteration. Some timings in seconds for the iterative algorithm and for the direct method are given in Table4 . All the calculations were programmed in Fortran 77, they were run on a Sun Ultra 10 workstation, and the times were measured by the DTIME instruction of the Sun Fortran Library.... In PAGE 27: ... All the calculations were programmed in Fortran 77, they were run on a Sun Ultra 10 workstation, and the times were measured by the DTIME instruction of the Sun Fortran Library. Separate measure- ments of the preliminary work and of a single iteration are reported in Table4 , the values of n and q being taken from Table 3. The table shows also the total times when the iterative algorithm is applied to two versions of Problem A, the positive value of c being 2 (1=n)1=d.... ..."