### TABLE V EXECUTION TIMES (MILLISECONDS) PERCENTILES AND MAX TIME Map Size 10th 25th 50th 75th 90th Max

### Table 1: Sample descriptive statistics: The 0th, the 10th, the 25th, the 50th, the 75th, the 90th and the 100th percentile of some characteristics of the cleaned data set used in the empirical analysis, namely the FTSE 100, the strike price, the option price (in points), the time to maturity (in days), the (annualized) interest rate (in %), the ex post (annualized) dividend yield (in %), the (annualized) implied volatility (in %). The data set consists of 65549 option contracts (33633 call options and 31916 put options).

in Risk-neutral Density Extraction from Option Prices: Improved Pricing with Mixture Density Networks

"... In PAGE 7: ... The final set consists of 65549 option contracts (33633 call options and 31916 put options). A brief description of the option contracts is given in Table1 . We report the 0th, the 10th, the 25th, the 50th, the 75th, the 90th and the 100th percentile12 of several sample characteristics.... In PAGE 7: ... The longest contract has a time to maturity of more than one year. In Table1 we also report the statistics of the risk-free interest rate13 and the ex post dividend yield of the FTSE 100 which are inputs to the option pricing models. Finally, the percentiles of the implied volatilities of all options are reported.... ..."

### Table 1. Derivative works of the Madonna title Nothing Fails in a commercial database.

"... In PAGE 4: ... This operation was performed for all of 20 query songs. We used textual title stem matches to identify ground truth derivative works, see Table1 . We recorded true posi- tives and false positives at each level of recall standardized into 10th-percentiles.... ..."

### Table 10: Beam noise levels (dB) at the 25th percentile for a clutter environment with own-ship speed equal to 5 knots.

### Table 2. Percentiles computed from the sample in Table 1 using linear interpolation (Deciles) and Vincent apos;s procedure (Vincentiles).

"... In PAGE 11: ..., the points on the y-axis are always the same (0, .1, .2, etc.), but the corresponding points on the x-axis are taken from Table2 . The points on the x-axis will vary from sample to sample.... In PAGE 13: ...e.g., Balota amp; Spieler, 1999; Spieler et al., 1996). We will use the Vincent histogram as a nonparametric density estimator without averaging. For the decile estimate, the deciles shown in Table2 are used as the real upper and lower limits of the histogram bars. The height of each bar is chosen so that the area of the bar is equal to :10, the relative frequency of observations falling within each region.... In PAGE 14: ... This bar will be centered at (682:6 ms + 705:4 ms)=2 = 694 ms, the midpoint of the interval between the rst and second vincentile, corresponding roughly to the 10th percentile. Using the vincentiles in Table2 , we construct 9 bars encompassing 90% of the sample. The upper and lower 5% of the data are not represented in the density estimate, although they contributed to the computation of the vincentiles.... ..."

### Table 2. Description of a subset of significant genes

"... In PAGE 5: ... As a comparison, a total of 694 genes were identified as significantly associated with the time to a recovery when their test statistics are compared to the 10th-percentiles of the normal distribution. A sample Gene Ontology for the selection of genes for which change in expression over time is associated with the time to res- piratory recovery is given in Table2 . The sample genes are grouped in those that exhibited positive association with the time to respi- ratory response, and those with a negative association.... ..."

### Table 3: Comparison of Model Fit and Prediction Statistics (N=958)a

"... In PAGE 15: ... This table reveals substantial variation in percentile values between males and females for some of the biomarkers. Table3 displays the R2, pseudo-R2, and area under the ROC curve statistics (both within and out-of-sample) for each model. Although the estimated model coefficients are not shown, Table 3 shows that only the summed GOM score (based on the 10th and 90th percentiles), the two-tailed 16-item count-based measure (based on the 10th and 90th percentiles), and the z-score index are significantly (p lt;0.... In PAGE 15: ... Table 3 displays the R2, pseudo-R2, and area under the ROC curve statistics (both within and out-of-sample) for each model. Although the estimated model coefficients are not shown, Table3 shows that only the summed GOM score (based on the 10th and 90th percentiles), the two-tailed 16-item count-based measure (based on the 10th and 90th percentiles), and the z-score index are significantly (p lt;0.05) associated with all five health outcomes in their respective models (first, seventh, and ninth columns of Table 3, respectively).... In PAGE 15: ... Although the estimated model coefficients are not shown, Table 3 shows that only the summed GOM score (based on the 10th and 90th percentiles), the two-tailed 16-item count-based measure (based on the 10th and 90th percentiles), and the z-score index are significantly (p lt;0.05) associated with all five health outcomes in their respective models (first, seventh, and ninth columns of Table3 , respectively). Conversely, the one-tailed 16-item count-based measure (using quartiles) and the conventional measure used in much of the existing literature (the one-tailed, 10-item measure based on quartiles) are significantly (p lt;0.... ..."

### Table 2: 10th: 10th root compression and mean normalization.

"... In PAGE 3: ...s 23.5% (Table 1). Simply applying 10th root compression and mean normalization reduced the average word error rate to 15.4% ( Table2 ). Applying quantile equalization further im- proves the result leading to an error rate of 13.... ..."

### Table 1. Distribution of properties of BaliBase data sets Value dN/dS Kappa

2008

"... In PAGE 6: ...pl (Cartwright 2005) was used to estimate lambda, the ratio of indels to base substitutions (Fig. 1, Table1 ). Kappa, lambda and the dN/dS ratio are thus the criteria by which these real data sets can be compared with EvolveAGene 3 simulated data sets.... In PAGE 6: ... Properties of simulated data sets Twenty of the BaliBase data sets were selected as models upon which to base EvolveAGene 3 simulations. Data sets were selected to include those with between 10 and 142 sequences, and to roughly represent the range between the 90th percentile and 10th percentile of the dN/dS, kappa and lambda distributions ( Table1 ). Ten simulations were based on each of the 20 BaliBase models.... ..."