%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 12 0 R 13 0 R ] /Type /Pages /Count 10 >> endobj 2 0 obj << /Subject (Neural Information Processing Systems http\072\057\057nips\056cc\057) /Publisher (Curran Associates\054 Inc\056) /Language (en\055US) /Created (2018) /EventType (Poster) /Description-Abstract (In many real\055world learning tasks\054 it is hard to directly optimize the true performance measures\054 meanwhile choosing the right surrogate objectives is also difficult\056 Under this situation\054 it is desirable to incorporate an optimization of objective process into the learning loop based on weak modeling of the relationship between the true measure and the objective\056 In this work\054 we discuss the task of objective adaptation\054 in which the learner iteratively adapts the learning objective to the underlying true objective based on the preference feedback from an oracle\056 We show that when the objective can be linearly parameterized\054 this preference based learning problem can be solved by utilizing the dueling bandit model\056 A novel sampling based algorithm DL\1362M is proposed to learn the optimal parameter\054 which enjoys strong theoretical guarantees and efficient empirical performance\056 To avoid learning a hypothesis from scratch after each objective function update\054 a boosting based hypothesis adaptation approach is proposed to efficiently adapt any pre\055learned element hypothesis to the current objective\056 We apply the overall approach to multi\055label learning\054 and show that the proposed approach achieves significant performance under various multi\055label performance measures\056) /Producer (PyPDF2) /Title (Preference Based Adaptation for Learning Objectives) /Date (2018) /ModDate (D\07220190219003510\05508\04700\047) /Published (2018) /Type (Conference Proceedings) /firstpage (7828) /Book (Advances in Neural Information Processing Systems 31) /Description (Paper accepted and presented at the Neural Information Processing Systems Conference \050http\072\057\057nips\056cc\057\051) /Editors (S\056 Bengio and H\056 Wallach and H\056 Larochelle and K\056 Grauman and N\056 Cesa\055Bianchi and R\056 Garnett) /Author (Yao\055Xiang Ding\054 Zhi\055Hua Zhou) /lastpage (7837) >> endobj 3 0 obj << /Type /Catalog /Pages 1 0 R >> endobj 4 0 obj << /Parent 1 0 R /Contents [ 14 0 R ] /Type /Page /Resources 15 0 R /MediaBox [ 0 0 612 792 ] >> endobj 5 0 obj << /Parent 1 0 R /Contents [ 66 0 R ] /Resources 67 0 R /MediaBox [ 0 0 612 792 ] /Annots 78 0 R /Type /Page >> endobj 6 0 obj << /Parent 1 0 R /Contents [ 93 0 R ] /Resources 94 0 R /MediaBox [ 0 0 612 792 ] /Annots 95 0 R /Type /Page >> endobj 7 0 obj << /Parent 1 0 R /Contents [ 106 0 R ] /Resources 107 0 R /MediaBox [ 0 0 612 792 ] /Annots 116 0 R /Type /Page >> endobj 8 0 obj << /Parent 1 0 R /Contents [ 132 0 R ] /Resources 133 0 R /MediaBox [ 0 0 612 792 ] /Annots 138 0 R /Type /Page >> endobj 9 0 obj << /Parent 1 0 R /Contents [ 143 0 R ] /Resources 144 0 R /MediaBox [ 0 0 612 792 ] /Annots 150 0 R /Type /Page >> endobj 10 0 obj << /Parent 1 0 R /Contents [ 158 0 R ] /Resources 159 0 R /MediaBox [ 0 0 612 792 ] /Annots 210 0 R /Type /Page >> endobj 11 0 obj << /Parent 1 0 R /Contents [ 212 0 R ] /Resources 213 0 R /MediaBox [ 0 0 612 792 ] /Annots 214 0 R /Type /Page >> endobj 12 0 obj << /Parent 1 0 R /Contents [ 223 0 R ] /Type /Page /Resources 224 0 R /MediaBox [ 0 0 612 792 ] >> endobj 13 0 obj << /Parent 1 0 R /Contents [ 225 0 R ] /Type /Page /Resources 226 0 R /MediaBox [ 0 0 612 792 ] >> endobj 14 0 obj << /Length 3140 /Filter /FlateDecode >> stream xڍYY~ϯG<lǰb[5C/wz俧1KY]U]]UN"vѮ<>LX=L;*zf˻ߨBxw3YT;*lww-۫
}cj౪vGY7UY_yչd7QZͪ"8rS|#&~128