%PDF-1.3 1 0 obj << /Kids [ 4 0 R 5 0 R 6 0 R 7 0 R 8 0 R 9 0 R 10 0 R 11 0 R ] /Type /Pages /Count 8 >> endobj 2 0 obj << /Subject (Neural Information Processing Systems http\072\057\057nips\056cc\057) /Publisher (Curran Associates) /Language (en\055US) /Created (2008) /Description-Abstract (In multi\055task learning several related tasks are considered simultaneously\054 with the hope that by an appropriate sharing of information across tasks\054 each task may benefit from the others\056 In the context of learning linear functions for supervised classification or regression\054 this can be achieved by including a priori information about the weight vectors associated with the tasks\054 and how they are expected to be related to each other\056 In this paper\054 we assume that tasks are clustered into groups\054 which are unknown beforehand\054 and that tasks within a group have similar weight vectors\056 We design a new spectral norm that encodes this a priori assumption\054 without the prior knowledge of the partition of tasks into groups\054 resulting in a new convex optimization formulation for multi\055task learning\056 We show in simulations on synthetic examples and on the iedb MHC\055I binding dataset\054 that our approach outperforms well\055known convex methods for multi\055task learning\054 as well as related non convex methods dedicated to the same problem\056) /Producer (Python PDF Library \055 http\072\057\057pybrary\056net\057pyPdf\057) /Title (Clustered Multi\055Task Learning\072 A Convex Formulation) /Date (2008) /Type (Conference Proceedings) /firstpage (745) /Book (Advances in Neural Information Processing Systems 21) /Description (Paper accepted and presented at the Neural Information Processing Systems Conference \050http\072\057\057nips\056cc\057\051) /Editors (D\056 Koller and D\056 Schuurmans and Y\056 Bengio and L\056 Bottou) /Author (Laurent Jacob\054 Jean\055philippe Vert\054 Francis R\056 Bach) /lastpage (752) >> endobj 3 0 obj << /Type /Catalog /Pages 1 0 R >> endobj 4 0 obj << /Contents 12 0 R /Rotate 0 /Resources << /ExtGState << /GS0 13 0 R >> /Font << /T1_6 14 0 R /T1_7 18 0 R /T1_4 22 0 R /T1_5 26 0 R /T1_2 31 0 R /T1_3 32 0 R /T1_0 37 0 R /T1_1 39 0 R >> /ProcSet [ /PDF /Text ] >> /CropBox [ 0 0 612 792 ] /Parent 1 0 R /MediaBox [ 0 0 612 792 ] /Type /Page >> endobj 5 0 obj << /Contents 41 0 R /Rotate 0 /Resources << /ExtGState << /GS0 13 0 R >> /Font << /T1_6 42 0 R /T1_7 22 0 R /T1_4 26 0 R /T1_5 14 0 R /T1_2 37 0 R /T1_3 32 0 R /T1_0 39 0 R /T1_1 45 0 R /T1_8 18 0 R /T1_9 46 0 R /T1_10 51 0 R /T1_11 55 0 R /T1_12 58 0 R /T1_13 61 0 R /T1_14 65 0 R >> /ProcSet [ /PDF /Text ] >> /CropBox [ 0 0 612 792 ] /Parent 1 0 R /MediaBox [ 0 0 612 792 ] /Type /Page >> endobj 6 0 obj << /Contents 69 0 R /Rotate 0 /Resources << /ExtGState << /GS0 13 0 R >> /Font << /T1_6 18 0 R /T1_7 51 0 R /T1_4 32 0 R /T1_5 26 0 R /T1_2 14 0 R /T1_3 22 0 R /T1_0 45 0 R /T1_1 39 0 R /T1_8 46 0 R /T1_9 58 0 R /T1_10 61 0 R /T1_11 65 0 R >> /ProcSet [ /PDF /Text ] >> /CropBox [ 0 0 612 792 ] /Parent 1 0 R /MediaBox [ 0 0 612 792 ] /Type /Page >> endobj 7 0 obj << /Contents 70 0 R /Rotate 0 /Resources << /ExtGState << /GS0 13 0 R >> /Font << /T1_6 18 0 R /T1_7 51 0 R /T1_4 14 0 R /T1_5 46 0 R /T1_2 32 0 R /T1_3 22 0 R /T1_0 26 0 R /T1_1 39 0 R /T1_8 61 0 R /T1_9 45 0 R /T1_10 58 0 R /T1_11 65 0 R /T1_12 42 0 R /T1_13 37 0 R >> /ProcSet [ /PDF /Text ] >> /CropBox [ 0 0 612 792 ] /Parent 1 0 R /MediaBox [ 0 0 612 792 ] /Type /Page >> endobj 8 0 obj << /Contents 71 0 R /Rotate 0 /Resources << /ExtGState << /GS0 13 0 R >> /Font << /T1_6 14 0 R /T1_7 61 0 R /T1_4 18 0 R /T1_5 22 0 R /T1_2 26 0 R /T1_3 32 0 R /T1_0 37 0 R /T1_1 39 0 R /T1_8 46 0 R /T1_9 45 0 R /T1_10 42 0 R /T1_11 58 0 R /T1_12 65 0 R /T1_13 51 0 R /T1_14 72 0 R /T1_15 75 0 R >> /ProcSet [ /PDF /Text ] >> /CropBox [ 0 0 612 792 ] /Parent 1 0 R /MediaBox [ 0 0 612 792 ] /Type /Page >> endobj 9 0 obj << /Contents 78 0 R /Rotate 0 /Resources << /ExtGState << /GS0 13 0 R >> /Font << /T1_6 18 0 R /T1_7 51 0 R /T1_4 26 0 R /T1_5 46 0 R /T1_2 32 0 R /T1_3 22 0 R /T1_0 39 0 R /T1_1 14 0 R /T1_8 61 0 R /T1_9 75 0 R /T1_10 37 0 R /T1_11 65 0 R /T1_12 45 0 R /T1_13 79 0 R /T1_14 42 0 R >> /ProcSet [ /PDF /Text ] >> /CropBox [ 0 0 612 792 ] /Parent 1 0 R /MediaBox [ 0 0 612 792 ] /Type /Page >> endobj 10 0 obj << /Contents 82 0 R /Rotate 0 /Resources << /ColorSpace << /CS0 83 0 R >> /ExtGState << /GS0 13 0 R >> /Font << /T1_4 45 0 R /T1_5 37 0 R /T1_2 32 0 R /T1_3 84 0 R /T1_0 39 0 R /T1_1 14 0 R >> /ProcSet [ /PDF /Text /ImageB /ImageI ] >> /CropBox [ 0 0 612 792 ] /Parent 1 0 R /MediaBox [ 0 0 612 792 ] /Type /Page >> endobj 11 0 obj << /Contents 85 0 R /Rotate 0 /Resources << /ExtGState << /GS0 13 0 R >> /Font << /T1_4 26 0 R /T1_5 45 0 R /T1_2 37 0 R /T1_3 32 0 R /T1_0 39 0 R /T1_1 14 0 R >> /ProcSet [ /PDF /Text ] >> /CropBox [ 0 0 612 792 ] /Parent 1 0 R /MediaBox [ 0 0 612 792 ] /Type /Page >> endobj 12 0 obj << /Length 3090 /Filter /FlateDecode >> stream HWk۶+Vc! gg:رMԵ7͇]OKQX6E|x=^zDM/X p\sZYh;*20-ltX|wz/Eu";mcCBzn=8խfRxq0Y4[$I4,[,p*uzjuà?b.ʃ]8hf6VQ.V&W6tXP,o?m2eto̓K(jAǺ*u/o㙫<'8/`LlU'1#erTc}1
ލ;S18X*Y0alciZ-?[_DpT0n{U`3x
3LTdsEzpp?LpYr[f͜Z_Z\t$g} }QF tZY_+`^4ʐ9XYlb='}5%*ޑ}fSZ&;B &Q@;{3ʶ髍'TEEڱd6c5H,Gl~Î`ײiZ`$E#2еZ" !={Nھ1h!-$TMk{CR,F?MWZͨ*x'=@b5E16%m'2xp3W,+Rؘ
uv2mCdJRk' :1
"9N7bD,divbŚ>nA7