%PDF-1.4
%
1 0 obj
<>
endobj
2 0 obj
<>stream
2012-02-14T21:51:53-05:00
2012-02-15T02:51:35Z
2012-02-14T21:51:53-05:00
Microsoft PowerPoint
uuid:91f6889d-1626-e843-9d0b-a61686901c9b
uuid:9c33545c-a0df-2f46-a525-0235e00ccc3c
application/pdf
2012-01-31_probabilityDecisions.ppt
Erik Sudderth
Adobe Mac PDF Plug-in
1
B
http://ns.adobe.com/xap/1.0/mm/
xmpMM
XMP Media Management Schema
internal
UUID based identifier for specific incarnation of a document
InstanceID
URI
http://www.aiim.org/pdfa/ns/id/
pdfaid
PDF/A ID Schema
internal
Part of PDF/A standard
part
Integer
internal
Amendment of PDF/A standard
amd
Text
internal
Conformance level of PDF/A standard
conformance
Text
endstream
endobj
5 0 obj
[<>]
endobj
3 0 obj
<>
endobj
7 0 obj
<>
endobj
8 0 obj
<>
endobj
9 0 obj
<>
endobj
10 0 obj
<>
endobj
26 0 obj
<>/ExtGState<>/Font<>/ProcSet[/PDF/Text/ImageC/ImageI]/XObject<>>>/Type/Page>>
endobj
27 0 obj
<>/ExtGState<>/Font<>/ProcSet[/PDF/Text]>>/Type/Page>>
endobj
28 0 obj
<>/ExtGState<>/Font<>/ProcSet[/PDF/Text/ImageC]/XObject<>>>/Type/Page>>
endobj
29 0 obj
<>/ExtGState<>/Font<>/ProcSet[/PDF/Text]>>/Type/Page>>
endobj
30 0 obj
<>/ExtGState<>/Font<>/ProcSet[/PDF/Text/ImageC]/XObject<>>>/Type/Page>>
endobj
31 0 obj
<>/ExtGState<>/Font<>/ProcSet[/PDF/Text]>>/Type/Page>>
endobj
32 0 obj
<>/ExtGState<>/Font<>/ProcSet[/PDF/Text/ImageC]/XObject<>>>/Type/Page>>
endobj
66 0 obj
<>stream
/CS0 cs 1 1 1 scn
/GS0 gs
36 36 720 540 re
f*
BT
/CS1 cs 0 0.392 0.886 scn
/TT0 1 Tf
40 0 0 40 249.5215 516.4865 Tm
(Cross-Validation )Tj
0 0 0 scn
/C2_0 1 Tf
24 0 0 24 67.2 453.2749 Tm
<003E>Tj
/CS0 cs 0 0 0 scn
/C2_1 1 Tf
<0001>Tj
/CS1 cs 0 0 0 scn
/TT0 1 Tf
0.557 0 Td
(Divide training data into )Tj
-0.016 -1.167 Td
(K equal-sized )Tj
1 0 0 scn
/TT1 1 Tf
6.337 0 Td
(folds )Tj
0 0 0 scn
/C2_0 1 Tf
-6.879 -1.208 Td
<003E>Tj
/CS0 cs 0 0 0 scn
/C2_1 1 Tf
<0001>Tj
/CS1 cs 0 0 0 scn
/TT0 1 Tf
0.557 0 Td
[(T)37(rain on K-1 folds, )]TJ
-0.016 -1.208 Td
(evaluate on remainder )Tj
/C2_0 1 Tf
-0.542 -1.208 Td
<003E>Tj
/CS0 cs 0 0 0 scn
/C2_1 1 Tf
<0001>Tj
/CS1 cs 0 0 0 scn
/TT0 1 Tf
0.557 0 Td
(Pick model with best )Tj
-0.016 -1.208 Td
(average performance )Tj
0 -1.167 TD
(across K trials )Tj
ET
q
351 249.5 405 248.5 re
W n
q
/GS1 gs
405.8108063 0 0 249.3107758 350.5945892 249.0946045 cm
/Im0 Do
Q
Q
BT
0 0.392 0.886 scn
/TT2 1 Tf
28 0 0 28 67.2 217.7749 Tm
(How many folds? )Tj
/C2_0 1 Tf
24 0 0 24 73.2 179.15 Tm
<003E>Tj
/CS0 cs 0 0.392 0.886 scn
/C2_1 1 Tf
<0001>Tj
/CS1 cs 0 0.392 0.886 scn
/TT1 1 Tf
0.557 0 Td
(Bias: )Tj
0 0 0 scn
/TT0 1 Tf
[(T)111(oo few)55(, and ef)18(fective training dataset much smaller )]TJ
0 0.392 0.886 scn
/C2_0 1 Tf
-0.557 -1.167 Td
<003E>Tj
/CS0 cs 0 0.392 0.886 scn
/C2_1 1 Tf
<0001>Tj
/CS1 cs 0 0.392 0.886 scn
/TT1 1 Tf
0.557 0 Td
[(V)37(ariance: )]TJ
0 0 0 scn
/TT0 1 Tf
[(T)111(oo many)74(, and test performance estimates noisy )]TJ
0 0.392 0.886 scn
/C2_0 1 Tf
-0.557 -1.208 Td
<003E>Tj
/CS0 cs 0 0.392 0.886 scn
/C2_1 1 Tf
<0001>Tj
/CS1 cs 0 0.392 0.886 scn
/TT1 1 Tf
0.557 0 Td
(Cost: )Tj
0 0 0 scn
/TT0 1 Tf
(Must run training algorithm once per fold \(parallelizable\) )Tj
0.427 0.89 0.016 scn
/C2_0 1 Tf
-0.557 -1.208 Td
<003E>Tj
/CS0 cs 0.427 0.89 0.016 scn
/C2_1 1 Tf
<0001>Tj
/TT1 1 Tf
0.557 0 Td
(Practical rule of thumb: )Tj
/CS1 cs 0 0 0 scn
/TT0 1 Tf
(5-fold or 10-fold cross-validation )Tj
1 0 0 scn
/C2_0 1 Tf
-0.557 -1.208 Td
<003E>Tj
/CS0 cs 1 0 0 scn
/C2_1 1 Tf
<0001>Tj
/CS1 cs 1 0 0 scn
/TT1 1 Tf
0.557 0 Td
(Theoretically troubled: )Tj
0 0 0 scn
/TT0 1 Tf
10.394 0 Td
(Leave-one-out cross-validation, K=N )Tj
ET
endstream
endobj
67 0 obj
<>/Filter/FlateDecode/Height 615/Length 94529/Name/X/Subtype/Image/Type/XObject/Width 1001>>stream
HoT3sb 6DAbPFH،:3g6J$[>(UPT~/2s<#T/Os=Gzys04CrL6P
iA"2[C7!Ȇ2a
#fÈ,_\ʹ2
M7R^XKD=L&ɘ"%U먪FR)Yc˱XFc뺦ixbQ$)Ib*d2X}ku%\fטDFI,,,&AB?oxkXt`ѝa{?KHHCn
p zstGnO%'3+, َQE4jgѱѽc5DWq!-3Y
sX,6;;+I5Ci,H)
ҵqNJ"tʊjfC