### Table 1. Performance Comparison on Ro- bust04

### Table 4.1 The values of the programmable bias and the corresponding minimum relative ro- bustness.

2007

### Table 1 shows that, for each method, there is at least one example for which it has the best performance, or close to the best. Therefore, it seems natural to ask the question of robustness. That is, how well a particular method m per- forms on average in situations that are most favorable to other procedures. Following Friedman [6], we capture ro- bustness by computing the ratio bm of its error rate em and the smallest error rate over all methods being compared in a particular example: bm =

2002

"... In PAGE 4: ... Although the variances in the first problem are most differ- ent between the two classes on the high coordinates, their means now are most separated in the lower coordinates, so that all variables contain substantial discriminating infor- mation. The second column of Table1 shows the results for this problem. 3.... In PAGE 4: ... The distribution of the bm values for each method m over all the examples, therefore, seems to be a good indicator con- cerning its robustness. Table1 : Average classification error rates for simulated data. Ex1 Ex2 Ex3.... ..."

Cited by 42

### Table 2: Results of Experiment 1, designed to test ro- bustness over time, repeatability and make a comparison with odometry. x denotes the true x-coordinate, x is the error in the tracking units estimate, xod error in odom- etry based estimate, etc ... The total time for the tests is given along with the theoretical distance traveled and the average speed.

1998

"... In PAGE 7: ... Even better performance would have been achieved by actively controlling the direc- tion in which the sensor was looking. Table2 shows that the repeatability was fairly good since the robot returned to almost the same position. The odometry was helpless when the platform drifted for about 30 during each test.... ..."

Cited by 1

### Table 1: Three commonly used M-estimators. 3.3 The Local Stage The aim of this stage is to reduce the in uence of er- roneous sections of the features: to perform this task, the residual error ri of curve Ci is computed by a ro- bust function of the distances fdi;jg1 j l0 i. We could take ri equal to the median of the fdi;jg, but once

"... In PAGE 4: ... These functions are very e cient, but are not suited to cases where the presence of outliers in the data is too large (experimentally, it must be kept below approximately 20%). Table1 lists three commonly used functions and their derivative. Among these estimators, some are more restrictive than others: when Tukey apos;s in uence function is null for residuals larger than a threshold c, Cauchy apos;s re- mains larger than zero while decreasing, whereas Hu- ber apos;s remains constant, equal to c.... In PAGE 5: ... Despite the bad accuracy of the model, the result is visually con- vincing. In order the reader to be aware of the parts of the curve which are less taken into account in the computation, we have drawn in black the points for which the residual is greater than c (c is de ned in Table1 ). Roughly speaking, these points are the ones for which the weight in the computation is decreased because their residual is too large.... ..."

### Table 2 presents the standard deviation relative to the value of every affine invariant under a 0:05 standard devi- ation noise and 5% of missing data, respectively. Due to the way we compute the affine invariants, by applying the inverse of the intrinsic reference system, the invariant ro- bustness is related to the reference system robustness. The two last columns of Table 2 show how well separated the scatter invariants are under noise and missing data.

"... In PAGE 5: ... Table2 . Percentage of error in invariants under 0:05 noise and under 5% missing data, respectively.... ..."

### Table 3: Global Phoneme Set [Worldbet notation] sounds [n], [d], [m], and [b], whereas others are not,as for example the phoneme [g], which is frequent in Ko- rean but extremely rare in Spanish. In the rst case sharing the data results in language independent ro- bust models of [n], [d], [m], and [b] and reduces the number of parameter of the nal system. In the latter case the less frequent phonemes like the Spanish [g] would bene t from sharing the training data across the languages, since more reliable estimation for this model can be achieved. Sounds belonging to only one language like [Nq] for Japanese or the apped [r(] for Spanish help solving the language identi cation prob- lem because these sounds are reliable predictors for one language.

2000

"... In PAGE 3: ... For eight languages this global set consists of 145 phoneme categories. Table3 shows the global phoneme set for ve languages in World- bet notation. About half of the set consists of mono- phonemes belonging to only one language, the other half is shared across at least two languages.... ..."

Cited by 1

### Table 4: Comparison of the results for the fourth problem (disjoint feasible region). RY = Runarsson amp; Yao [31], KM = Koziel amp; Michalewicz [22]. N.A. = Not Available.

2002

"... In PAGE 24: ... We can observe a very ro- bust behavior of the proposed technique in this example. The best solution of our approach is compared in Table4 . Note that the approach proposed by Koziel amp; Michalewicz [22] required 1,400,000 evaluations of the tness function to produce the result shown in Table 4.... ..."

Cited by 4

### Table 1: Statistical profiles considering Constraint FT (numbers are in per cent)

2006

"... In PAGE 3: ... The test sets have 100% ro- bust path delay coverage and are not compacted. Table1 quotes the results considering only Constraint FT (functional transition). The first four columns contain the name of the circuit, number of PIs, FFs and TPs generated by TIP.... ..."

Cited by 1

### Table 2: Results for synset translation using glosses Test word

"... In PAGE 4: ...-associated-word correlation matrices. Table2 lists the results. The experimental results reveal that the method is not ro- bust; glosses are very short, and a few associated words deter- mine the translation of a word in a synset.... ..."