📄 pres2.ps
字号:
32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font525 1191 314 (Non-parametric) 314 SB32 0 0 67 67 0 0 0 60 /Times-Roman /font32 ANSIFont font300 1310 51 (8.) 51 SB375 1310 769 (Combining type 1 classifiers) 769 SB32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font450 1388 1631 (One thing that every classifier type can do is output a label \(i.e. output the specific) 1631 SB450 1447 1601 (class that the sample X is \(apparently\) associated with. Classifiers such as Bayes) 1601 SB450 1506 1597 (can be reduced to this level by simply discarding information. Thus, for the sake) 1597 SB450 1565 1220 (of generality, this type of combination is the primary concern.) 1220 SB32 0 0 67 67 0 0 0 60 /Times-Roman /font32 ANSIFont font300 1625 51 (9.) 51 SB375 1625 186 (Voting) 186 SB32 0 0 42 42 0 0 0 42 /Symbol fontgs 469 3231 0 0 CB450 1706 19 (\267) 19 SBgr32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font525 1703 1539 (This is the simplest way to combine classifiers. It works by consensus \(i.e. all) 1539 SB525 1762 1446 (experts must decide the same thing\). This will cause the rejection rate to) 1446 SB525 1821 207 (skyrocket.) 207 SB32 0 0 42 42 0 0 0 42 /Symbol fontgs 469 3231 0 0 CB450 1883 19 (\267) 19 SBgr32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font525 1880 1448 (Alternative method simply make sure that there is a majority between the) 1448 SB525 1939 157 (experts.) 157 SB32 0 0 67 67 0 0 0 60 /Times-Roman /font32 ANSIFont font300 1999 85 (10.) 85 SB450 1999 480 (Confusion Matrix) 480 SB32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font450 2077 1472 (The confusion matrix is simply a way to assess the "quality" of a classifier.) 1472 SB450 2136 187 (Example:) 187 SB450 2254 1625 (We are given a test sample size of 10: 5 from class 1 and 5 from class 2. Suppose) 1625 SB450 2313 979 (that we have a classifier that performs as follows:) 979 SB600 2431 877 (Classifies 3 of the class 1 samples as class 1.) 877 SB600 2490 877 (Classifies 1 of the class 1 samples as class 2.) 877 SB600 2549 638 (Rejects 1 of the class 1 samples.) 638 SB600 2608 877 (Classifies 4 of the class 2 samples as class 2.) 877 SB600 2667 877 (Classifies 1 of the class 2 samples as class 1.) 877 SB600 2726 638 (Rejects 0 of the class 2 samples.) 638 SB450 2844 621 (The resulting confusion matrix:) 621 SB32 0 0 42 42 0 0 0 38 /Times-Italic /font31 ANSIFont font300 3069 912 (CPSC 599.14, Oct 18, 1993 A.D. Paul Schermerhorn) 912 SB1 #Cstatusdict begin /manualfeed false store endEJ RS%%PageTrailer%%PageResources: font Symbol%%+ font Times-Bold%%+ font Times-Italic%%+ font Times-Roman%%Page: 4 4%%PageResources: (atend)SS0 0 25 11 798 1100 300 SM32 0 0 50 50 1 0 0 45 /Times-Italic /font31 ANSIFont font0 0 0 fCgs 207 59 613 176 CB636 178 22 (e) 22 SBgr32 0 0 50 50 1 0 0 45 /Times-Roman /font32 ANSIFont fontgs 207 59 613 176 CB658 178 123 (\(X\)=1) 123 SBgr32 0 0 50 50 1 0 0 45 /Times-Italic /font31 ANSIFont fontgs 208 59 821 176 CB843 178 22 (e) 22 SBgr32 0 0 50 50 1 0 0 45 /Times-Roman /font32 ANSIFont fontgs 208 59 821 176 CB865 178 123 (\(X\)=2) 123 SBgr32 0 0 50 50 1 0 0 45 /Times-Italic /font31 ANSIFont fontgs 362 59 1030 176 CB1053 178 22 (e) 22 SBgr32 0 0 50 50 1 0 0 45 /Times-Roman /font32 ANSIFont fontgs 362 59 1030 176 CB1075 178 255 (\(X\)=rejected) 255 SBgr32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont fontgs 184 59 428 235 CB450 237 33 (C) 33 SBgr32 0 0 50 50 0 0 0 45 /Times-Italic /font31 ANSIFont fontgs 184 59 428 235 CB483 250 42 (1:) 42 SBgr32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont fontgs 207 59 613 235 CB704 237 25 (3) 25 SBgrgs 208 59 821 235 CB913 237 25 (1) 25 SBgrgs 362 59 1030 235 CB1199 237 25 (1) 25 SBgrgs 184 59 428 294 CB450 296 33 (C) 33 SBgr32 0 0 50 50 0 0 0 45 /Times-Italic /font31 ANSIFont fontgs 184 59 428 294 CB483 309 42 (2:) 42 SBgr32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont fontgs 207 59 613 294 CB704 296 25 (1) 25 SBgrgs 208 59 821 294 CB913 296 25 (4) 25 SBgrgs 362 59 1030 294 CB1199 296 25 (0) 25 SBgr450 414 1470 (The training samples should theoretically have the same distribution as the) 1470 SB450 473 304 (expected input.) 304 SB32 0 0 67 67 0 0 0 60 /Times-Roman /font32 ANSIFont font300 592 85 (11.) 85 SB450 592 162 (Bayes) 162 SB32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font450 670 1561 (From the confusion matrix, the "prior knowledge of the experts" can be found:) 1561 SB1059 733 45 (P\() 45 SB32 0 0 50 50 0 0 0 51 /Symbol fontgs 1138 3231 0 0 CB1104 727 34 (w) 34 SBgr32 0 0 50 50 0 0 0 45 /Times-Italic /font31 ANSIFont font1138 746 38 (1 ) 38 SB32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font1176 733 314 (| e\(X\)=j\) = 0.75) 314 SB1059 798 45 (P\() 45 SB32 0 0 50 50 0 0 0 51 /Symbol fontgs 1138 3231 0 0 CB1104 792 34 (w) 34 SBgr32 0 0 50 50 0 0 0 45 /Times-Italic /font31 ANSIFont font1138 811 38 (2 ) 38 SB32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font1176 798 314 (| e\(X\)=j\) = 0.80) 314 SB450 859 1168 (One can ensure that the classifier output is independent by:) 1168 SB32 0 0 42 42 0 0 0 42 /Symbol fontgs 469 3231 0 0 CB450 921 19 (\267) 19 SBgr32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font525 918 668 (Training them with different data.) 668 SB32 0 0 42 42 0 0 0 42 /Symbol fontgs 469 3231 0 0 CB450 980 19 (\267) 19 SBgr32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font525 977 815 (Using mutually exclusive feature subsets.) 815 SB32 0 0 67 67 0 0 0 60 /Times-Roman /font32 ANSIFont font300 1096 85 (12.) 85 SB450 1096 455 (Dempster-Shafer) 455 SB32 0 0 50 50 1 0 0 45 /Times-Roman /font32 ANSIFont font450 1174 823 (Bayes: the chance the a hypothesis is true) 823 SB450 1233 1397 (D-S: the chance that the evidence supports the truth of the hypothesis.) 1397 SB32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font450 1292 1032 (One gets some evidence, and it has carried with it a ) 1032 SB32 0 0 50 50 0 0 0 45 /Times-Italic /font31 ANSIFont font1482 1292 334 (probability mass) 334 SB32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font1816 1292 116 (. The) 116 SB450 1351 1458 (combined mass function is then updated, from which the belief function is) 1458 SB450 1410 1605 (computed. The maximum of all the beliefs is \(hopefully\) the class that X belongs) 1605 SB450 1469 54 (to.) 54 SB32 0 0 50 50 0 0 0 45 /Times-Italic /font31 ANSIFont font451 1589 61 (bel) 61 SB540 1589 31 (A) 31 SB704 1589 36 (m) 36 SB762 1589 31 (B) 31 SB32 0 0 29 29 0 0 0 27 /Times-Italic /font31 ANSIFont font641 1655 18 (B) 18 SB685 1655 18 (A) 18 SB32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font515 1589 17 (\() 17 SB570 1589 17 (\)) 17 SB740 1589 17 (\() 17 SB793 1589 17 (\)) 17 SB32 0 0 50 50 0 0 0 51 /Symbol font599 1583 27 (=) 27 SB32 0 0 29 29 0 0 0 29 /Symbol font660 1653 21 (\315) 21 SB32 0 0 75 75 0 0 0 76 /Symbol font644 1569 53 (\345) 53 SB32 0 0 50 50 0 0 0 45 /Times-Italic /font31 ANSIFont font453 1751 36 (m) 36 SB507 1751 33 (C) 33 SB612 1751 36 (m) 36 SB712 1751 36 (m) 36 SB816 1751 36 (m) 36 SB1017 1751 36 (m) 36 SB1091 1751 31 (A) 31 SB1162 1751 36 (m) 36 SB1237 1751 31 (B) 31 SB32 0 0 29 29 0 0 0 27 /Times-Italic /font31 ANSIFont font936 1817 18 (A) 18 SB979 1817 18 (B) 18 SB1016 1817 19 (C) 19 SB32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font489 1751 17 (\() 17 SB543 1751 17 (\)) 17 SB1066 1751 17 (\() 17 SB1121 1751 17 (\)) 17 SB1215 1751 17 (\() 17 SB1268 1751 17 (\)) 17 SB32 0 0 50 50 0 0 0 51 /Symbol font572 1745 27 (=) 27 SB666 1745 38 (\305) 38 SB776 1745 27 (=) 27 SB892 1745 27 (=) 27 SB1144 1745 13 (\327) 13 SB32 0 0 29 29 0 0 0 29 /Symbol font953 1815 22 (\307) 22 SB998 1815 16 (=) 16 SB32 0 0 75 75 0 0 0 76 /Symbol font957 1731 53 (\345) 53 SB32 0 0 29 29 0 0 0 26 /Times-Roman /font32 ANSIFont font647 1770 14 (1) 14 SB749 1770 14 (2) 14 SB850 1770 28 (12) 28 SB1052 1770 14 (1) 14 SB1198 1770 14 (2) 14 SB32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font450 1852 1555 (Xu et. al then makes use of the fact that this case is somewhat degenerate, and) 1555 SB450 1911 1041 (turns an exp-time algorithm into an O\(M\) algorithm.) 1041 SB32 0 0 67 67 0 0 0 60 /Times-Roman /font32 ANSIFont font300 2030 85 (13.) 85 SB450 2030 197 (Results) 197 SB32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font450 2108 1039 (This is a good trick to use to increase "the numbers") 1039 SB1 lc1 lj0 0 0 pC0 2 SP717 2281 M 371 0 1 PPSn32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font454 2249 55 (Re) 55 SB780 2217 55 (Re) 55 SB792 2288 42 (%) 42 SB882 2288 55 (Re) 55 SB32 0 0 50 50 0 0 0 45 /Times-Italic /font31 ANSIFont font516 2249 156 (liability) 156 SB842 2217 189 (cognition) 189 SB953 2288 136 (jection) 136 SB32 0 0 50 50 0 0 0 51 /Symbol font676 2243 27 (=) 27 SB844 2282 27 (-) 27 SB32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font717 2288 75 (100) 75 SB32 0 0 42 42 0 0 0 42 /Symbol fontgs 469 3231 0 0 CB450 2353 19 (\267) 19 SBgr32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font525 2350 564 (D-S is best and most robust.) 564 SB32 0 0 42 42 0 0 0 42 /Symbol fontgs 469 3231 0 0 CB450 2412 19 (\267) 19 SBgr32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font525 2409 476 (Bayes is a close second.) 476 SB32 0 0 42 42 0 0 0 42 /Symbol fontgs 469 3231 0 0 CB450 2471 19 (\267) 19 SBgr32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font525 2468 1357 (All the methods yield significantly better results than the best expert.) 1357 SB32 0 0 42 42 0 0 0 42 /Symbol fontgs 469 3231 0 0 CB450 2530 19 (\267) 19 SBgr32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font525 2527 1116 (They never tested it for statistical significance, however.) 1116 SB32 0 0 42 42 0 0 0 42 /Symbol fontgs 469 3231 0 0 CB450 2589 19 (\267) 19 SBgr32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font525 2586 541 (What's wrong with graphs?) 541 SB32 0 0 67 67 0 0 0 60 /Times-Roman /font32 ANSIFont font300 2705 85 (14.) 85 SB450 2705 306 (Conclusion) 306 SB32 0 0 42 42 0 0 0 42 /Symbol fontgs 469 3231 0 0 CB450 2786 19 (\267) 19 SBgr32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font525 2783 651 (Not the last word on the subject.) 651 SB32 0 0 42 42 0 0 0 42 /Symbol fontgs 469 3231 0 0 CB450 2845 19 (\267) 19 SBgr32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font525 2842 626 (Towards Plug'n Play classifiers.) 626 SB32 0 0 42 42 0 0 0 42 /Symbol fontgs 469 3231 0 0 CB450 2904 19 (\267) 19 SBgr32 0 0 50 50 0 0 0 45 /Times-Roman /font32 ANSIFont font525 2901 370 (Computing trends.) 370 SB32 0 0 42 42 0 0 0 38 /Times-Italic /font31 ANSIFont font300 3069 912 (CPSC 599.14, Oct 18, 1993 A.D. Paul Schermerhorn) 912 SB1 #Cstatusdict begin /manualfeed false store endEJ RS%%PageTrailer%%PageResources: font Symbol%%+ font Times-Italic%%+ font Times-Roman%%Trailerend%%Pages: 4% TrueType font name key:% MSTT310000 = % MSTT310000 = % MSTT310000 = % MSTT310000 = % MSTT310000 = % MSTT310000 = % MSTT310000 = % MSTT31c355 = 2f2fDTimes New RomanF00000043000002bc0000% MSTT31c362 = 2f2fDTimes New RomanF00000032000001900000% MSTT31c36f = 2f2fDTimes New RomanF00000043000001900000% MSTT31c37c = 2f2fDTimes New RomanF00000032000002bc0000% MSTT31c389 = 2f2fDTimes New RomanF00000032000001900001% MSTT31c396 = 2f2fDTimes New RomanF0000002a000001900001% MSTT31c3a3 = 2f2fDTimes New RomanF0000003a000002bc0000% MSTT31c3b0 = 2f2fDTimes New RomanF00000007000001900000% MSTT31c3bd = 2f2fDTimes New RomanF0000001d000001900001% MSTT31c3ca = 2f2fDTimes New RomanF0000001d000001900000% MSTT31c3d7 = 2f2fDTimes New RomanF00000006000001900000%%DocumentSuppliedResources: procset Win35Dict 3 1%%DocumentNeededResources: font Symbol%%+ font Times-Bold%%+ font Times-Italic%%+ font Times-Roman%%EOF⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -