%PDF-1.3
1 0 obj
<<
/Kids [ 4 0 R 5 0 R 6 0 R 7 0 R ]
/Type /Pages
/Count 4
>>
endobj
2 0 obj
<<
/Subject (Neural Information Processing Systems http\072\057\057nips\056cc\057)
/Publisher (Curran Associates)
/Language (en\055US)
/Created (2011)
/Description-Abstract (Unsupervised feature learning has been shown to be effective at learning representations that perform well on image\054 video and audio classification\056 However\054 many existing feature learning algorithms are hard to use and require extensive hyperparameter tuning\056 In this work\054 we present sparse filtering\054 a simple new algorithm which is efficient and only has one hyperparameter\054 the number of features to learn\056 In contrast to most other feature learning methods\054 sparse filtering does not explicitly attempt to construct a model of the data distribution\056 Instead\054 it optimizes a simple cost function \055\055 the sparsity of L2\055normalized features \055\055 which can easily be implemented in a few lines of MATLAB code\056 Sparse filtering scales gracefully to handle high\055dimensional inputs\054 and can also be used to learn meaningful features in additional layers with greedy layer\055wise stacking\056 We evaluate sparse filtering on natural images\054 object classification \050STL\05510\051\054 and phone classification \050TIMIT\051\054 and show that our method works well on a range of different modalities\056)
/Producer (Python PDF Library \055 http\072\057\057pybrary\056net\057pyPdf\057)
/Title (Spotlight Slides\072 Sparse Filtering)
/Date (2011)
/Type (Conference Proceedings)
/firstpage (1125)
/Book (Advances in Neural Information Processing Systems 24)
/Description (Paper accepted and presented at the Neural Information Processing Systems Conference \050http\072\057\057nips\056cc\057\051)
/Editors (J\056 Shawe\055Taylor and R\056S\056 Zemel and P\056L\056 Bartlett and F\056 Pereira and K\056Q\056 Weinberger)
/Author (Jiquan Ngiam\054 Zhenghao Chen\054 Sonia A\056 Bhaskar\054 Pang W\056 Koh\054 Andrew Y\056 Ng)
/lastpage (1133)
>>
endobj
3 0 obj
<<
/Type /Catalog
/Pages 1 0 R
>>
endobj
4 0 obj
<<
/Parent 1 0 R
/Contents 8 0 R
/Tabs /S
/Resources <<
/ExtGState <<
/GS6 9 0 R
/GS7 10 0 R
/GS12 11 0 R
/GS5 12 0 R
/GS19 13 0 R
/GS18 14 0 R
>>
/XObject <<
/Image20 15 0 R
/Image13 16 0 R
/Image15 17 0 R
/Image8 18 0 R
/Image14 20 0 R
>>
/Font <<
/F2 21 0 R
/F1 25 0 R
>>
/ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ]
>>
/Group <<
/Type /Group
/S /Transparency
/CS /DeviceRGB
>>
/MediaBox [ 0 0 720 540 ]
/Type /Page
/StructParents 0
>>
endobj
5 0 obj
<<
/Parent 1 0 R
/Contents 29 0 R
/Tabs /S
/Resources <<
/XObject <<
/Image30 30 0 R
/Image8 18 0 R
/Image35 31 0 R
>>
/Pattern <<
/P38 32 0 R
/P29 33 0 R
/P31 34 0 R
/P32 35 0 R
/P33 36 0 R
/P34 37 0 R
/P36 38 0 R
/P37 39 0 R
>>
/ExtGState <<
/GS28 40 0 R
/GS7 10 0 R
/GS12 11 0 R
>>
/Font <<
/F2 21 0 R
/F3 41 0 R
/F1 25 0 R
/F4 48 0 R
>>
/ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ]
>>
/Group <<
/Type /Group
/S /Transparency
/CS /DeviceRGB
>>
/MediaBox [ 0 0 720 540 ]
/Type /Page
/StructParents 1
>>
endobj
6 0 obj
<<
/Parent 1 0 R
/Contents 56 0 R
/Tabs /S
/Resources <<
/XObject <<
/Image51 57 0 R
/Image8 18 0 R
>>
/ExtGState <<
/GS7 10 0 R
/GS12 11 0 R
>>
/Font <<
/F2 21 0 R
/F1 25 0 R
/F5 59 0 R
>>
/ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ]
>>
/Group <<
/Type /Group
/S /Transparency
/CS /DeviceRGB
>>
/MediaBox [ 0 0 720 540 ]
/Type /Page
/StructParents 2
>>
endobj
7 0 obj
<<
/Parent 1 0 R
/Contents 66 0 R
/Tabs /S
/Resources <<
/ExtGState <<
/GS7 10 0 R
/GS12 11 0 R
/GS5 12 0 R
>>
/XObject <<
/Image55 67 0 R
/Image57 69 0 R
/Image8 18 0 R
>>
/Font <<
/F2 21 0 R
/F1 25 0 R
/F6 70 0 R
>>
/ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ]
>>
/Group <<
/Type /Group
/S /Transparency
/CS /DeviceRGB
>>
/MediaBox [ 0 0 720 540 ]
/Type /Page
/StructParents 3
>>
endobj
8 0 obj
<<
/Length 1692
/Filter /FlateDecode
>>
stream
xWn7}SDwAƉ$M.6erWڕr
Æɝ+9sP}=_/ӉJ*HkDJ8m3|x$ɓWgN,RH_>N~Nw'BίHcuy~[Qbu\v^7K8eǝ+D=W+ѓDOn/$ya#IS/5C||=E~ bYL5S Qڡ}9Όt6jcN#ۚLuStu5˄ /81-TpOGe?S5U͈q ]ifvКin`pcA=Wj-sD)FwjFD?v}m7OS^:Bg%2b%DNA5kZ "$
dܓ$i'nX0"=6ar!o7)ux-[,Q: Np)iB97d;cNםSȍ:O:ڝݘhE`~ 16(viWj>
@ȲiTF/QP媗֔ uWDZ<W73dkmJ r VeoW(34k8lc:@dR&Y|r72,',k$T>{,fsVK˦7:ll.W<Z$:_M(Jm@*J.\:̄Vǹaw>ĸH!С!++c|uT#J~t:|գ='0B:茜n@=ء\#,KC١ME7ti5gpN\K1;8U|8sp75Ȗ\oNٌ\@hb)k\nth5'LIpR۳ZlX6;ط17%=W`\&`Hw.Go;lG_/I>K ^HyH`@\[nxvxld5c?t=hר6rY; ۾}쎪Q}UОpeԎ}UÕ;^U N3G;~Pe
v^F @h\TlM1krT"P5Qd8o +Re
Sy ueyKrLuNե18#mѨ
6\}ucYL+19o_ "PUBmjz|@3Sn^g&gvy9O\|w\LQpZVuct"*ԊAÉ}UcQIĭ7Q>$ [67P1vRSeZ
QmeR\El$:oS]0|~*s EC
"v|`01
#< g3"
endstream
endobj
9 0 obj
<<
/ca 0.80000
/Type /ExtGState
/BM /Normal
>>
endobj
10 0 obj
<<
/CA 1
/Type /ExtGState
/BM /Normal
>>
endobj
11 0 obj
<<
/CA 0.65882
/Type /ExtGState
/BM /Normal
>>
endobj
12 0 obj
<<
/ca 1
/Type /ExtGState
/BM /Normal
>>
endobj
13 0 obj
<<
/CA 0.81176
/Type /ExtGState
/BM /Normal
>>
endobj
14 0 obj
<<
/ca 0.16078
/Type /ExtGState
/BM /Normal
>>
endobj
15 0 obj
<<
/Filter /FlateDecode
/Interpolate false
/Length 12973
/ColorSpace /DeviceRGB
/Type /XObject
/BitsPerComponent 8
/Height 270
/Width 269
/Subtype /Image
>>
stream
xi%iYn2:〸 (ж 6B^{-3,w[U]K/M/,N-(8O019;|x3~nW=<ޕ7YQ
'BYitlZ-g,L&53,\<V?悁U%JgQ[JGQurmF/UjDU@ LGFo
ơcTX
Zr1[xua!~ϱ2ILR&)I$e29L^K[Ә%.&Į-**UᔎؗI8X"V66/c/ iiw|ebSxw-׃ֵ$c.qE͈l#G(JI*gxSB|53^]SX%>L
*`3hQ[%i_ϣ_Vh
I$e2ILR&)ɁeRK8zjo(.^w;okm