"... In PAGE 4: ... We also measured the \cross-product quot; of the regis- ter operand sources on the two sides: each instruction can be characterized by its choices on the two input multiplexors. Table3 shows the average bypass behav- ior of the integer benchmarks on our baseline model. It records for each (A-side, B-side) pair the average fraction of instruction executions that used that pair.... In PAGE 4: ... This matches the one-dimensional view of Table 2 in the obvious way. The northwest corner of Table3 describes instruc- tion executions that needed no bypassing, either be- cause they had no register operands (like jmp instruc- tions), or because both of their operands came from the register le itself, without bene t of bypassing. In these programs, about half of all instructions executed are in that upper corner.... In PAGE 5: ...Table3 : Average bypass usage in integer benchmarks (normalized to instruction count) A-side B-side operand operand No Reg Reg File Ex Mem Wb Totals No Reg 0.039 0.... ..."

"... In PAGE 2: ... For others, the anatomical con- text is defined only to a relatively coarse level of granular- ity by microdissecting mesoscopic brain regions [18-20]. Associated with these brain atlas projects and reference data sets are software systems for browsing the images in a 3D context (see Table1 ), though only a subset of these supports reslicing the 3D data sets to produce novel 2D views independent of the orientation in which the 2D images were originally acquired (e.g.... ..."

"... In PAGE 5: ... Table2 : Results for different view cells. The ef ciency of the different stages of the algorithm is shown in Table 3.... In PAGE 7: ... So in total, the algorithm complexity is between O(klogk+logklogn) and O(k2 logn). We provide experimental evidence about the derived com- plexity bounds by measuring the dependence of the the size of the line space BSP tree and the total number of funnel vis- ibility tests on the number of visible occluders for the scenes and tests presented in Table2 . The measurements are de- picted in Figure 6.... ..."

"... In PAGE 9: ... Binary features typically capture part relationships such as dis- tances, angles, and also include boundary relationships (see below). Typical examples of these feature types are shown in Table1 (centre column) and those used in this implementation are shown in Table 1 (right column). The furthest column groups features into di erent types: unary curvatures(U.... In PAGE 9: ... Binary features typically capture part relationships such as dis- tances, angles, and also include boundary relationships (see below). Typical examples of these feature types are shown in Table 1 (centre column) and those used in this implementation are shown in Table1 (right column). The furthest column groups features into di erent types: unary curvatures(U.... In PAGE 9: ... At run time, however, new data needs to be equally transformed using the statistics generated during training. Table1 shows our reformu- lated computations of feature types aimed at capturing the fundamental geometry/topology of the objects and their views, and in ways which are as view-independent as possible. Further, we have used features which are real- or integer- valued and not simply present/absent, as used in previous implementations.... In PAGE 23: ... Speci cally, we have studied the performance of the system with respect to the number of rules, the cost func- tion: distance (K-Means) versus minimum entropy, and nally, the feature types.In accord with the model procedures, we have implemented segmentation, feature extraction ( Table1 ) and rule generation-weight estimation on both sets of training data. Examples of rule bounds are shown in Table 2, using the minimum entropy technique, just for the unary features (U.... In PAGE 23: ... Examples of rule bounds are shown in Table 2, using the minimum entropy technique, just for the unary features (U.x) as outlined in Table1 . Since z-scores were used (see Eqn.... In PAGE 28: ...D U.B Figure 7: Best Classi cation performance for each of the ve best feature types (see Table1 ) for the Chess Set(CS) classi cation problem (representative entropy values are shown). Refer to Ta- ble 1 for feature de nitions and brief descriptions.... ..."

"... In PAGE 5: ... Especially at low bit- rates, computing times can be increased by 40% (cf. Table3 ), but the absolute difference is not substantial. For lossless coding, the increase in computing time drops to about 12%.... In PAGE 7: ... The downscaling from factor 2 to factor 4 has to be done at the client side. Table3 compares the data transmission ti- mes required for a 1024x1024 image, and a rect- angular fish eye view of that image with both the classic wavelet decomposition and the desired method with independent x and y resolutions.... In PAGE 8: ... Compared to the gains when applying the scheme (cf. Table3 ), these costs can easily be justified. If the wavelet filter supports perfect reconstruc- tion, this property is maintained under the new decomposition scheme.... ..."

"... In PAGE 7: ... Since the closest point search could be performed independently for all surfaces, the method is ideally suited for a parallel implementation. Figure 6 and Table1 illustrate some results of the multi-view registration algorithm. (a) (b) (c) (d) (e) (f) Figure 6.... ..."

"... In PAGE 14: ... Table3 : The average mean square errors between the projections derived from the Fourier Pro- jection Theorem and those from spatial domain summing, for various viewing angles. The color values are normalized to the range between 0 and 255.... ..."

"... In PAGE 4: ... To measure e ectiveness, the list of documents whose full text was viewed during each search was extracted from the experiment logs. The aspectual recall of this list can be cal- culated using either the judgement of the independent NIST assessors, as shown in Table6 , or the judgement of the exper- imental subjects, as shown in Table 7. The assessors apos; judge- ment provides an objective assessment of the quality of the documents that were chosen for viewing, whereas the subjects apos; judgement re ects their own concept of document apos;s relevance to the topic in terms of their own understanding of the infor- mation need.... ..."

"... In PAGE 4: ... To measure e ectiveness, the list of documents whose full text was viewed during each search was extracted from the experiment logs. The aspectual recall of this list can be cal- culated using either the judgement of the independent NIST assessors, as shown in Table 6, or the judgement of the exper- imental subjects, as shown in Table7 . The assessors apos; judge- ment provides an objective assessment of the quality of the documents that were chosen for viewing, whereas the subjects apos; judgement re ects their own concept of document apos;s relevance to the topic in terms of their own understanding of the infor- mation need.... ..."