### Table 6: The increase in parallel time after random weight perturbation.

"... In PAGE 24: ... We bound the weight perturbation to a random amount between -20% and +20%. Then we observe the increase in parallel time shown in Table6 . The increase in parallel time is virtually none for small number of processors and very modest for a larger number of processors.... ..."

### Table 3: Discretization errors on randomly perturbed meshes using 1 ux unknown per curved face.

in The mimetic finite difference discretization of diffusion problem on unstructured polyhedral meshes

2005

Cited by 2

### Table 4: Discretization errors on randomly perturbed meshes using 4 ux unknowns per curved face.

in The mimetic finite difference discretization of diffusion problem on unstructured polyhedral meshes

2005

Cited by 2

### Table 3. Five-fold cross-validation on all 5217 objects. Right: Modest random search (5 random perturbations, 20 restarts); Left: Extensive random search (100 random perturbations; 100 random restarts).

1996

"... In PAGE 4: ... Notice that the ANN produces superior performance. Table3 outlines the overall performance of OC1 using a ve-fold cross-validation experiment. Cross-validation stud- ies of this sort are known to produce reliable estimates of accuracy while avoiding \optimistic quot; bias (Salzberg et al.... ..."

Cited by 2

### Table 1. Mean, spread and error scores for two values of the annual cycle Tac for ensembles based only on random perturbations to the initial field of PUMA

2002

"... In PAGE 4: ... 4 Results Ensembles generated by only perturbing the initial field of PUMA do not show relevant differences for the forecast scores values as IR is changed. Table1 shows the forecast scores2 obtained for two values of the annual cycle Tac for 2For this experiment, the control trajectory was considered to be... ..."