Tools
Sorted by:
Try your query at:
 |
 |
 |
 |
 |
 |
Results 21 - 30
of
22,422
Table 2. Communication and Meta-data Sizes for opt Versions
"... In PAGE 10: ... We next focus on the overhead of meta-data that needs to be sent with optimized messages that are comprised of dis- joint rectangular sections. Table2 presents the total com- munication volume (in Mega Bytes), i.e.... In PAGE 10: ...ages. The size of meta-data is less than 0.03% of the to- tal communication volume in all cases. The data presented in Table2 clearly establishes that the overhead of sending meta-data with optimized messages is negligible. Table 3 shows the total communication volume for the full versions.... In PAGE 10: ... Table 3 shows the total communication volume for the full versions. These numbers can be compared against the total communication volume shown for opt versions in Table2 to see the reduction in communication volume achieved by our runtime technique. For mg-vscope, with medium query size, the reduction in communication vol- ume is 9%, 25%, and 52% on 2, 4, and 8 processors, re- spectively.... ..."
Table 2. Communication and Meta-data Sizes for opt Versions
"... In PAGE 10: ... We next focus on the overhead of meta-data that needs to be sent with optimized messages that are comprised of dis- joint rectangular sections. Table2 presents the total com- munication volume (in Mega Bytes), i.e.... In PAGE 10: ...ages. The size of meta-data is less than 0.03% of the to- tal communication volume in all cases. The data presented in Table2 clearly establishes that the overhead of sending meta-data with optimized messages is negligible. Table 3 shows the total communication volume for the full versions.... In PAGE 10: ... Table 3 shows the total communication volume for the full versions. These numbers can be compared against the total communication volume shown for opt versions in Table2 to see the reduction in communication volume achieved by our runtime technique. For mg-vscope, with medium query size, the reduction in communication vol- ume is 9%, 25%, and 52% on 2, 4, and 8 processors, re- spectively.... ..."
Table 4 Hybrid tagging performance (all in %). Experiments are performed on three di erent document sets as explained in the text. The full tagger designates the POSTAG system with all the morphological processing capabilities. The tagger-a is a version without employing post error-correction rules. The tagger-b is a more degraded version which does not utilize our unknown morpheme guessing capability but treats all unknown morphemes as nouns (typical method of estimation). The tagger-c is an even more deteriorated version which rejects all unknown morphemes as tagging failures. The performance drops as we degrade the version from the full tagger to tagger-a, b, c respectively.
2002
"... In PAGE 18: ... This document set contains about 14% unknown morphemes (mainly technical jargon). Table4 shows our tagging performance for these three doc- ument sets. This experiment shows e ciency of our unknown morpheme handling and guessing techniques since we can con rm the sharp performance drops between tagger-a and tagger-b.... ..."
Cited by 7
Table 1: Features for each heapsort version, Exper- iments 1 and 2.
"... In PAGE 3: ... The full-featured version of the heapsort AV used for the experiment is shown in Figure 1. The features associated with each of the ve versions is shown in Table1 . The avail- able features indicated in the table include a \next quot; button to control speed of the tutorial, a \back quot; button to back up one logical step in the presentation, an example given by the AV, the ability for the user to enter his or her own example, pseudocode, and a question guide, repsectively.... ..."
Table 1: Features for each heapsort version, Exper- iments 1 and 2.
"... In PAGE 3: ... The full-featured version of the heapsort AV used for the experiment is shown in Figure 1. The features associated with each of the ve versions is shown in Table1 . The avail- able features indicated in the table include a \next quot; button to control speed of the tutorial, a \back quot; button to back up one logical step in the presentation, an example given by the AV, the ability for the user to enter his or her own example, pseudocode, and a question guide, repsectively.... ..."
Table 3: Compilers used for the version control simulations.
1993
"... In PAGE 16: ... Specifically, the performance of the p+1-bit full directory [Censier1978] is compared to the compiler-directed version control coherence mechanism [Cheong1989]. The range of performance of the version control scheme is estimated using three different levels of compiler technology, as summarized in Table3 . The simple compiler has imprecise memory disambiguation in that it maintains one version number (i.... ..."
Cited by 10
Table 3: Compilers used for the version control simulations.
1993
"... In PAGE 16: ... Specifically, the performance of the p+1-bit full directory [Censier1978] is compared to the compiler-directed version control coherence mechanism [Cheong1989]. The range of performance of the version control scheme is estimated using three different levels of compiler technology, as summarized in Table3 . The simple compiler has imprecise memory disambiguation in that it maintains one version number (i.... ..."
Cited by 10
Table 2. Error on training and testing sets for wavelet and DCT features (default zero threshold for all machines). Each set includes an equal number of examples of cover and stego images. Due to the high computational load associated with training on the wavelet features, in this submission we only provide results for a smaller database cosisting of 800 training and 320 testing images. Results for the full database will appear in the nal version.
"... In PAGE 7: ... The composition of training examples for each particular SVM is given in Table 1. The SVM parameters together with errors on the training and testing sets are shown in Table2 . We conclude that for grayscale JPEG images, the DCT features perform better than wavelet features.... ..."
Table 1: timings in seconds for the Brusselator Note that the RADAU5 and VODE trials were not performed for the largest values of N due to the inordinate length of time required, and that the su x b denotes banded versions. It is clear from this table how the two matrix-free methods clearly outperform the two direct Jacobian methods. For the larger values of M, this di erence is two to three orders of magnitude. This is in addition to only allocating O(M) bytes of memory as opposed to the O(M2) required for the full-matrix versions. The RADpk code was also parallelised in accordance with the description in section 3 on the Brusselator problem and was run with several di erent values for N with 1, 2 and 3 processors on a shared memory SGI Power Challenge sited at the University of Queensland. The times are given in Table 2 table below. M = 2N2
1997
Cited by 1
Table 2: Program Versions: This table lists the versions of the programs we used.
2000
"... In PAGE 9: ... The results in this sec- tion are preliminary and are intended as a proof of concept and not an evaluation of the full potential of the technique. We compare the results of the latest publicly available ver- sions (listed in Table2 ) of pathchar [7], clink [4], pchar [11], and nettimer. pathchar, clink, and pchar implement the one-packet technique described earlier.... ..."
Cited by 161
Results 21 - 30
of
22,422
