### Table 4: Performance of Approximate Distance Computation Algorithms (10% Relative Error) based on di erent BV types.

1999

Cited by 58

### Table 4: Performance of Approximate Distance Computation Algorithms #2810#25 Relative

1999

Cited by 58

### Table 1: Overview of previous and new results related to the computation of minimum-distance spanning trees and minimum distance-approximating spanning trees.

"... In PAGE 4: ...n Section 5). Finally, we extend results by Gerey et al. [GLJ78] and Hassin amp; Tamir [HT95] concerning the minimization of the distance matrix DT under various matrix norms, integrating the previous findings into a unifying framework in Theorem 1. (See Table1 for an overview of previous and new results.) This work is structured as follows: Section 2 gives an overview of previous results related to this work and new results obtained, followed by some definitions and easy observations in Section 3.... ..."

### Table 4: Computation/approximation trade o in the NN-link construction among image regions. The distance to a neighboring point is approximated to within (1+ ) times the actual distance. =0 indicates the exact k-NN computation. Elapse time: average wall clock time for one nearest neighbor search. Speedup: the ratio of elapse time, with respect to the time of sequential search (SS). Error: the percentage of mistakes made by approximation in the k nearest neighbors. The symbol \- quot; means zero error.

in Abstract

"... In PAGE 23: ... ANN estimates the distance to a nearest neighbor up to (1+ ) times the actual distance: = 0 means exact search, no approximation; bigger values give rougher estimation. Table4 lists the average wall clock time to compute the top 10 neighbors of a region in the 10 Corel image sets of our image captioning experiments. Compared to the sequential search, the speedup of using a spatial method increases from 12.... ..."

### Table 1: Results showing the stability of the reflective symmetry descriptor at different approximations. Stability was measured by computing the C4 BE -difference between the full reflective symmetry descriptor and the approximating descriptor, and dividing by the twice standard deviation in descriptors. Timing results include the time for rasterization and computation of the exponentially decaying Euclidean Distance Transform.

2003

Cited by 29

### Table 1: Distance computation matrix

1998

"... In PAGE 6: ... Various schemes can be used for approximating distance between any two members that are not exchanging timestamps. Table1 shows one such scheme for computing the distance from member a to another member b. In this scheme a local member computes a longer distance to a far away member than that computed by its 2In some cases, more than one representative might echo the local-session messages.... ..."

Cited by 8

### TABLE II COMPUTATIONAL TIME OF INITIAL (tinit) AND COVERING (tcover) RUNS OF MOPSO IN SECONDS. Dmin AND Dmax ARE THE MINIMUM AND MAXIMUM DISTANCES BETWEEN THE NON-DOMINATED SOLUTIONS OF THE APPROXIMATED PARETO-FRONT

2004

Cited by 5

### Table 1: Results showing the stability of the reflective symmetry descriptor at different approximations. Stability was measured by computing the a6a8a7 -difference between the full reflective symmetry descriptor and the approximating descriptor, and dividing by the twice standard deviation in descriptors. Timing results include the time for rasterization and computation of the exponentially decaying Euclidean Distance Transform.

2003

Cited by 29

### Table 2: Example computation for distance test. boak hank matt shaw

"... In PAGE 6: ... The absolute di erence in the log returns is found for the four extensions from the true extension, then these values are standardized to have mean 0 and variance 1. Table2 shows the computations for the rst series. The statistic that is the sum of the standardized distances of the submitted answers will be approximately normally distributed with a known variance.... ..."

### Table 1: Positioning Error. The average positioning error per workspace region is given in absolute distance and as a percentage of the 1.9m diameter workspace. The trained SOMs are used to approximate the regularized direct inverse functions. The resulting joint angles are used by the true forward kinematics function to obtain the actual position in the workspace, and the distance from the target computed. The averages are computed over 10,000 workspace points chosen at random, each with a random setting of the redundancy parameter.

1996

"... In PAGE 23: ...orkspace, with a maximum error of 16.7cm. The performance in W1 is the best, which is to be expected, given that it contains 170 20 = 3400 SOM nodes, but there is only small variation in positioning error across the regions. Table1 shows the average positioning error in centimeters for targets in each region. The averages were computed from a random (uniform) sample of 10,000 points in the workspace.... ..."

Cited by 7