"... In PAGE 13: ...99 and the num ber of iterations within SA to be 1,000,000. Table2 lists the re- sults of using different evaluation functions on the obtained solutions. For the weighted-sum objective function, we use the sam e set of weight values as in formula (29), and list the num - ber of archived non-dom inated solutions (see colum n 2) and the best solution under this evaluation function (see colum n 3).... In PAGE 14: ...Table 2. Results of the our proposed multi-objective approach after 1-hour runtime A ccording to the results in Table2 , we can see that our proposed approach is very prom ising in solving the m ulti-objective nurse scheduling problem . In terms of the solution quality evaluated by the sam e objective function, our approach performs similar to the IP-based VNS, and significantly improve the best results of the hybrid genetic algorithm and the hybrid VNS by 25.... ..."

"... In PAGE 6: ... Table 3: Results of Multimodal Experiments. Function SGA SGA with elite Tabu-GA Rastrigin 829/829/1 141/497/5 73/264/10 FMS- parameters -/-/0 126/282/2 143/608/10 Table4 shows the results of some multi-objective experiments, where Pareto is the number of Pareto optima and (n) is the tabu list size. While Ranking Selection GA gets six Pareto solutions on the frontier line of Pareto optima, Tabu-GA can get flexible and diverse Pareto solutions depending on a tabu list size.... ..."

"... In PAGE 11: ... Many advanced multi-objective evolutionary algorithms use some form of density dependent selection. Furthermore, nearly all techniques can be expressed in terms of density estimation, a classification is given in Table3 . We will make use of this as a further step towards a common framework of evolutionary multi-objective optimizers, and present the relevant enhancement of the unified model.... ..."

"... In PAGE 11: ... Many advanced multi-objective evolutionary algorithms use some form of density dependent selection. Furthermore, nearly all techniques can be expressed in terms of density estimation, a classification is given in Table3 . We will make use of this as a further step towards a common framework of evolutionary multi-objective optimizers, and present the relevant enhancement of the unified model.... ..."

"... In PAGE 6: ... The results are shown in Fig. 4, Table2 , and video meeting2 mcmc 500:avi. This sequence depicts four seated speakers in a more animated conversation (see SGT row in Table 2), with many turns and cases of overlapped speech.... In PAGE 6: ...eeting2. The results are shown in Fig. 4, Table 2, and video meeting2 mcmc 500:avi. This sequence depicts four seated speakers in a more animated conversation (see SGT row in Table2 ), with many turns and cases of overlapped speech. There are also two sources of visual clutter: the tex- tured background, and a fth walking person (not tracked) who enters and leaves the scene creating visual distraction by approaching the speakers.... In PAGE 7: ... Figures (a-c) correspond to frames 575, 860, and 909, respectively. covers in almost all cases, as shown by the SR, TR, and FT rows in Table2 . The combination of visual cues renders the tracker more robust: while the spatial structure obser- vations help in cases of uncertainty with respect to edge information (e.... ..."

"... In PAGE 35: ... The weighting numbers used in [12] are unknown and so this accounts for the variation in these parameters. Table3 presents a comparison of the performance (P), temperature (T), area (A), wire- length (W), and runtime of 4 di erent objective functions for the 2D and 3D cases. All data in this table are taken from the combined LP+SA approach.... In PAGE 37: ...Table3 is the pipeline depth and whitespace percentages for the various objective functions. Pipeline depth is calculated by adding in the number of ip ops inserted between the major stages of the basic simplescalar pipeline.... In PAGE 37: ... One can observe that there is a 15% reduction in IPC and a 22% reduction in temperature between the performance-only objective (0) and the highest weight hybrid objective (20) for the 3D case. As expected and also shown in Table3 the multi-layer oorplans increase both the temperature and IPC over the single layer oorplans. Also of note is that the highest thermal weight multi-layer oorplan has a temperature close to that of the lowest thermal weight single layer oorplan while achieving a higher IPC.... ..."

"... In PAGE 24: ... This is the reason why we run 30 times our multi-objective GA (denoted as GAm), compute the average result and standard deviation for each instance, and compare them to the ones obtained with our single-objective GA. Results are shown in Table4 . For each method, the best result in all runs, the average of the best results, the standard deviation, and the relative difference between the best known result and the average of the best results are given.... ..."

"... In PAGE 6: ... In this sequence, recorded with no visual background clutter, four seated speakers are engaged in a conversation and talk at a relaxed pace, tak- ing turns with little overlap, which occurs for instance when people laugh. The last row in Table1 (SGT ) indicates the proportion of time during which each person spoke in the sequence, as labeled in the speaking activity GT. Regarding visual tracking, the four objects were tracked with good quality and stably throughout the sequence for all runs (see SR, TR, and FT rows in Table 1, and video meeting1 mcmc 500:avi).... In PAGE 6: ... The last row in Table 1 (SGT ) indicates the proportion of time during which each person spoke in the sequence, as labeled in the speaking activity GT. Regarding visual tracking, the four objects were tracked with good quality and stably throughout the sequence for all runs (see SR, TR, and FT rows in Table1 , and video meeting1 mcmc 500:avi). The algorithm can handle partial visual self-occlusion (e.... In PAGE 6: ... With respect to speaking activity, our source localization method, combined with the AV calibration procedure, has shown to be able to estimate location reasonably well, and detect speaker turns with good accuracy and low latency, when people talk at the meeting table [6]. The audio ac- tivity inferred by the MCMC-PF preserves these properties for those segments where only one speaker takes the turn, while smoothing out very short speaker turns with the dy- namical model (see FS row in Table1 ). Although we use a 1www:idiap:ch= gatica=av-tracking-multiperson:html: a b c Figure 3: Multispeaker tracking results, meeting1.... In PAGE 6: ... To study the e ciency of the MCMC-PF, we compare it with a traditional joint multi-object PF, which uses IS in- stead of MCMC, while all other aspects and parameters of the lter remain xed. Results are computed using 20 runs, and are shown in Table1 and video meeting1 pf 500:avi. Clearly, our approach outperforms the traditional PF in both ability to track and estimation of the speaking sta- tus.... ..."