### Table 7: Forward-backward reestimation

### Table 3: Performance comparison for difierent models and parsing directions Model Forward Backward

"... In PAGE 2: ... The direction which parses from left to right is forward parsing, while the direction which parses from right to left is backward parsing. Table3 shows the results of our comparison, which show that backward parsing performed better than forward parsing for both SVM and CRF models, but there is no evidential difierence between the SVM models with difierent multiclass extensions. For all models, precision is substantially better than recall.... In PAGE 3: ... Table 4 shows the flnal test results of this model, as well as the flnal results of the unions of CRF with the two SVM models, as reported by the organizer. The results show that our simple ensemble model remarkably enhanced recall, with all recall results ranked in the top quartile, while precision results dropped slightly, compared with the results in Table3 . All f-score results were ranked in the top quartile, too.... ..."

### Table 4.11 Forward, Backward and Stepwise Selection Results for Model 1 Process Model Selected

2006

### Table 4.12 Forward, Backward and Stepwise Selection Results for Model 2 Process Model Selected

2006

### Table 4: Results of Forward and Backward Variable-Selection Procedure for Logit Analysis

2003

"... In PAGE 20: ...20 INSERT FIGURE 1 ABOUT HERE INSERT FIGURE 2 ABOUT HERE The results in Table4 show that the (forward as well as backward) variable-selection procedure results in almost identical models, i.e.... ..."

### Table 3: NF ; NB: observed number of ISR events in the forward and backward hemispheres ; Acounting FB

1997

"... In PAGE 12: ...symmetry by at most 0.02 in the energy region between 40 and 88 GeV. The asymmetry was also calculated, after correcting for the selection ine ciency as described in Section 6.1, by counting the number of events with positive and negative cos , since AFB = NF ? NB NF + NB (19) with NF = Z 1 0 dN d cos d cos NB = Z 0 ?1 dN d cos d cos : (20) Table3 shows the asymmetries obtained with the two methods as a function of ps0 up to 87 GeV. The raw asymmetries obtained with the t were corrected for the contamination by FSR events in a similar way as the polar angle distributions were corrected (see Sec- tion 6.... In PAGE 12: ...1). These corrected values are also shown in Table3 , and are displayed in Figure 6 together with the SM prediction for the Improved Born asymmetry. Figure 6 also shows the asymmetries measured by DELPHI near the Z0 peak (see [12,14]), after correction to Improved Born values.... In PAGE 12: ... For the systematic error, the error on the purity of the sample was taken into account. The numbers obtained are given as a function of ps0 in Table3 and are shown in Figure 7, together with the prediction of the Standard Model. The theoretical predictions include all electroweak radiative corrections, apart from the box diagrams which are very small... ..."

### Table 2: The performance of the parallel forward and backward substitutions algorithms. The run time (in seconds) is shown for 16, 32, 64, and 128 processors, for the forward and backward substitutions for each one of the nine factorizations of ILUT and ILUT*. The last row also shows the amount of time required by the parallel matrix-vector multiplication algorithm.

"... In PAGE 16: ... The last row also shows the amount of time required by the parallel matrix-vector multiplication algorithm. From Table2 we see that the amount of time required to solve the triangular systems increases as m increases since the triangular factors contain more elements. Also note that for fixed m, the time required also increases with decreasing t.... ..."

### Table 2: Results of the forward-backward pruning method.

1999

"... In PAGE 3: ...he word graph was 4.44 % at a WGD of 105.08. Then we used for- ward pruning on the one hand and forward-backward pruning on the other hand to reduce the size of the word graph with different values for fLat and fLat fb. The results are shown in Table 1 for the forward pruning and Table2 for the forward-backward prun- ing (FLat and FLat fb denote the logarithm of fLat and fLat fb). Comparing the results of the two pruning methods we see, that forward-backward pruning leads to smaller word graphs.... ..."

Cited by 10