diff --git "a/parse/train/SJe3HiC5KX/SJe3HiC5KX_middle.json" "b/parse/train/SJe3HiC5KX/SJe3HiC5KX_middle.json" new file mode 100644--- /dev/null +++ "b/parse/train/SJe3HiC5KX/SJe3HiC5KX_middle.json" @@ -0,0 +1,29384 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 106, + 79, + 431, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 78, + 431, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 78, + 431, + 97 + ], + "score": 1.0, + "content": "LEARNING FACTORIZED REPRESENTATIONS", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 98, + 387, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 98, + 387, + 117 + ], + "score": 1.0, + "content": "FOR OPEN-SET DOMAIN ADAPTATION", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 113, + 135, + 306, + 157 + ], + "lines": [ + { + "bbox": [ + 112, + 134, + 301, + 147 + ], + "spans": [ + { + "bbox": [ + 112, + 134, + 216, + 147 + ], + "score": 1.0, + "content": "Mahsa Baktashmotlagh", + "type": "text" + }, + { + "bbox": [ + 229, + 135, + 301, + 146 + ], + "score": 1.0, + "content": "Masoud Faraki∗", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 113, + 145, + 308, + 159 + ], + "spans": [ + { + "bbox": [ + 113, + 146, + 217, + 158 + ], + "score": 1.0, + "content": "University of Queensland", + "type": "text" + }, + { + "bbox": [ + 228, + 145, + 308, + 159 + ], + "score": 1.0, + "content": "Monash University", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 322, + 135, + 398, + 157 + ], + "lines": [ + { + "bbox": [ + 320, + 134, + 397, + 146 + ], + "spans": [ + { + "bbox": [ + 320, + 134, + 397, + 146 + ], + "score": 1.0, + "content": "Tom Drummond*", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 320, + 145, + 400, + 159 + ], + "spans": [ + { + "bbox": [ + 320, + 145, + 400, + 159 + ], + "score": 1.0, + "content": "Monash University", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 413, + 135, + 495, + 156 + ], + "lines": [ + { + "bbox": [ + 411, + 134, + 496, + 146 + ], + "spans": [ + { + "bbox": [ + 411, + 134, + 496, + 146 + ], + "score": 1.0, + "content": "Mathieu Salzmann", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 411, + 146, + 440, + 158 + ], + "spans": [ + { + "bbox": [ + 411, + 146, + 440, + 158 + ], + "score": 1.0, + "content": "EPFL", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + }, + { + "type": "title", + "bbox": [ + 278, + 186, + 333, + 198 + ], + "lines": [ + { + "bbox": [ + 276, + 185, + 336, + 200 + ], + "spans": [ + { + "bbox": [ + 276, + 185, + 336, + 200 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 143, + 210, + 468, + 375 + ], + "lines": [ + { + "bbox": [ + 141, + 210, + 469, + 222 + ], + "spans": [ + { + "bbox": [ + 141, + 210, + 469, + 222 + ], + "score": 1.0, + "content": "Domain adaptation for visual recognition has undergone great progress in the past", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 221, + 470, + 233 + ], + "spans": [ + { + "bbox": [ + 141, + 221, + 470, + 233 + ], + "score": 1.0, + "content": "few years. Nevertheless, most existing methods work in the so-called closed-set", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 232, + 470, + 245 + ], + "spans": [ + { + "bbox": [ + 141, + 232, + 470, + 245 + ], + "score": 1.0, + "content": "scenario, assuming that the classes depicted by the target images are exactly the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 140, + 241, + 469, + 257 + ], + "spans": [ + { + "bbox": [ + 140, + 241, + 469, + 257 + ], + "score": 1.0, + "content": "same as those of the source domain. In this paper, we tackle the more challeng-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 255, + 469, + 266 + ], + "spans": [ + { + "bbox": [ + 141, + 255, + 469, + 266 + ], + "score": 1.0, + "content": "ing, yet more realistic case of open-set domain adaptation, where new, unknown", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 265, + 469, + 278 + ], + "spans": [ + { + "bbox": [ + 141, + 265, + 469, + 278 + ], + "score": 1.0, + "content": "classes can be present in the target data. While, in the unsupervised scenario,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 276, + 469, + 288 + ], + "spans": [ + { + "bbox": [ + 141, + 276, + 469, + 288 + ], + "score": 1.0, + "content": "one cannot expect to be able to identify each specific new class, we aim to au-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 142, + 288, + 470, + 299 + ], + "spans": [ + { + "bbox": [ + 142, + 288, + 470, + 299 + ], + "score": 1.0, + "content": "tomatically detect which samples belong to these new classes and discard them", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 298, + 470, + 311 + ], + "spans": [ + { + "bbox": [ + 141, + 298, + 470, + 311 + ], + "score": 1.0, + "content": "from the recognition process. To this end, we rely on the intuition that the source", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 309, + 470, + 321 + ], + "spans": [ + { + "bbox": [ + 141, + 309, + 470, + 321 + ], + "score": 1.0, + "content": "and target samples depicting the known classes can be generated by a shared sub-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 321, + 469, + 331 + ], + "spans": [ + { + "bbox": [ + 141, + 321, + 469, + 331 + ], + "score": 1.0, + "content": "space, whereas the target samples from unknown classes come from a different,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 331, + 469, + 343 + ], + "spans": [ + { + "bbox": [ + 141, + 331, + 469, + 343 + ], + "score": 1.0, + "content": "private subspace. We therefore introduce a framework that factorizes the data into", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 342, + 469, + 353 + ], + "spans": [ + { + "bbox": [ + 141, + 342, + 469, + 353 + ], + "score": 1.0, + "content": "shared and private parts, while encouraging the shared representation to be dis-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 352, + 469, + 365 + ], + "spans": [ + { + "bbox": [ + 141, + 352, + 469, + 365 + ], + "score": 1.0, + "content": "criminative. Our experiments on standard benchmarks evidence that our approach", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 142, + 364, + 393, + 376 + ], + "spans": [ + { + "bbox": [ + 142, + 364, + 393, + 376 + ], + "score": 1.0, + "content": "outperforms the state of the art in open-set domain adaptation.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 16 + }, + { + "type": "title", + "bbox": [ + 108, + 394, + 206, + 407 + ], + "lines": [ + { + "bbox": [ + 105, + 393, + 208, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 208, + 410 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 419, + 505, + 507 + ], + "lines": [ + { + "bbox": [ + 105, + 418, + 505, + 432 + ], + "spans": [ + { + "bbox": [ + 105, + 418, + 505, + 432 + ], + "score": 1.0, + "content": "In many practical machine learning scenarios, the test samples are drawn from a different distribu-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 429, + 505, + 443 + ], + "spans": [ + { + "bbox": [ + 105, + 429, + 505, + 443 + ], + "score": 1.0, + "content": "tion from the training ones, due to varying acquisition conditions, such as different data sources,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 439, + 505, + 454 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 505, + 454 + ], + "score": 1.0, + "content": "illumination conditions and cameras, in the context of visual recognition. Over the years, great", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 451, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 505, + 466 + ], + "score": 1.0, + "content": "progress has been achieved to tackle this problem, known has the domain shift. In particular, many", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 463, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 463, + 505, + 475 + ], + "score": 1.0, + "content": "methods aim to align the source (i.e., training) and target (i.e., test) distributions by learning domain-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 474, + 505, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 505, + 486 + ], + "score": 1.0, + "content": "invariant embeddings (Pan et al., 2011; Gong et al., 2012; Fernando et al., 2013; Sun et al., 2016),", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 485, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 505, + 497 + ], + "score": 1.0, + "content": "the most recent approaches relying on deep networks (Ganin & Lempitsky, 2014; Long et al., 2015;", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 496, + 426, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 426, + 508 + ], + "score": 1.0, + "content": "Bousmalis et al., 2016; Tzeng et al., 2017; Long et al., 2016a; Yan et al., 2017).", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 107, + 513, + 505, + 677 + ], + "lines": [ + { + "bbox": [ + 106, + 512, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 106, + 512, + 505, + 525 + ], + "score": 1.0, + "content": "While effective, these methods work under the assumption that the source and target data contain ex-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 524, + 505, + 536 + ], + "spans": [ + { + "bbox": [ + 106, + 524, + 505, + 536 + ], + "score": 1.0, + "content": "actly the same classes. In practice, however, this assumption may easily be violated, as the target data", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 535, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 535, + 505, + 547 + ], + "score": 1.0, + "content": "will often contain additional classes that were not present within the source data. For example, when", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 545, + 505, + 558 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 505, + 558 + ], + "score": 1.0, + "content": "training a model to recognize office objects from images, as with the popular Office dataset (Saenko", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 104, + 554, + 505, + 570 + ], + "spans": [ + { + "bbox": [ + 104, + 554, + 505, + 570 + ], + "score": 1.0, + "content": "et al., 2010), one should still expect to see new objects, unobserved during training, when deploying", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 567, + 506, + 580 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 506, + 580 + ], + "score": 1.0, + "content": "the model in the real world. While one should not expect the model to recognize the specific class", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 578, + 506, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 506, + 591 + ], + "score": 1.0, + "content": "of such objects, at least in unsupervised domain adaptation where no target labels are provided, it", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 589, + 505, + 602 + ], + "spans": [ + { + "bbox": [ + 105, + 589, + 505, + 602 + ], + "score": 1.0, + "content": "would nonetheless be beneficial to identify these objects as unknown instead of misclassifying them.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 600, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 505, + 613 + ], + "score": 1.0, + "content": "This was the task addressed by Busto & Gall (2017) in their so-called open-set domain adaptation", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 611, + 506, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 611, + 506, + 624 + ], + "score": 1.0, + "content": "approach. This method aims to learn a mapping from the source samples to a subset of the target", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 104, + 621, + 506, + 636 + ], + "spans": [ + { + "bbox": [ + 104, + 621, + 506, + 636 + ], + "score": 1.0, + "content": "ones corresponding to those identified as coming from known classes. While reasonably effective,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 633, + 506, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 506, + 645 + ], + "score": 1.0, + "content": "this procedure involves alternatively solving for the mapping and the assignment of the samples to", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "score": 1.0, + "content": "known/unknown classes, which, as shown in our experiments, can be costly. Recently, Saito et al.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "(2018) introduced a deep learning framework for open-set domain adaptation, relying on adversarial", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 667, + 430, + 678 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 430, + 678 + ], + "score": 1.0, + "content": "training to separate the samples from the known classes from the unknown ones.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 108, + 683, + 502, + 705 + ], + "lines": [ + { + "bbox": [ + 105, + 682, + 506, + 696 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 506, + 696 + ], + "score": 1.0, + "content": "In this paper, we introduce a novel approach to open-set domain adaptation based on learning a", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 694, + 504, + 705 + ], + "spans": [ + { + "bbox": [ + 106, + 694, + 504, + 705 + ], + "score": 1.0, + "content": "factorized representation of the source and target data. In essence, we seek to model the samples", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 48.5 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 712, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 117, + 709, + 506, + 724 + ], + "spans": [ + { + "bbox": [ + 117, + 709, + 506, + 724 + ], + "score": 1.0, + "content": "∗This work was supported by the Australian Research Council Centre of Excellence for Robotic Vision", + "type": "text" + } + ] + }, + { + "bbox": [ + 106, + 721, + 222, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 222, + 732 + ], + "score": 1.0, + "content": "(project number CE14010006).", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 106, + 79, + 431, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 78, + 431, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 78, + 431, + 97 + ], + "score": 1.0, + "content": "LEARNING FACTORIZED REPRESENTATIONS", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 98, + 387, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 98, + 387, + 117 + ], + "score": 1.0, + "content": "FOR OPEN-SET DOMAIN ADAPTATION", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 113, + 135, + 306, + 157 + ], + "lines": [ + { + "bbox": [ + 112, + 134, + 301, + 147 + ], + "spans": [ + { + "bbox": [ + 112, + 134, + 216, + 147 + ], + "score": 1.0, + "content": "Mahsa Baktashmotlagh", + "type": "text" + }, + { + "bbox": [ + 229, + 135, + 301, + 146 + ], + "score": 1.0, + "content": "Masoud Faraki∗", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 113, + 145, + 308, + 159 + ], + "spans": [ + { + "bbox": [ + 113, + 146, + 217, + 158 + ], + "score": 1.0, + "content": "University of Queensland", + "type": "text" + }, + { + "bbox": [ + 228, + 145, + 308, + 159 + ], + "score": 1.0, + "content": "Monash University", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5, + "bbox_fs": [ + 112, + 134, + 308, + 159 + ] + }, + { + "type": "text", + "bbox": [ + 322, + 135, + 398, + 157 + ], + "lines": [ + { + "bbox": [ + 320, + 134, + 397, + 146 + ], + "spans": [ + { + "bbox": [ + 320, + 134, + 397, + 146 + ], + "score": 1.0, + "content": "Tom Drummond*", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 320, + 145, + 400, + 159 + ], + "spans": [ + { + "bbox": [ + 320, + 145, + 400, + 159 + ], + "score": 1.0, + "content": "Monash University", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 320, + 134, + 400, + 159 + ] + }, + { + "type": "text", + "bbox": [ + 413, + 135, + 495, + 156 + ], + "lines": [ + { + "bbox": [ + 411, + 134, + 496, + 146 + ], + "spans": [ + { + "bbox": [ + 411, + 134, + 496, + 146 + ], + "score": 1.0, + "content": "Mathieu Salzmann", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 411, + 146, + 440, + 158 + ], + "spans": [ + { + "bbox": [ + 411, + 146, + 440, + 158 + ], + "score": 1.0, + "content": "EPFL", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5, + "bbox_fs": [ + 411, + 134, + 496, + 158 + ] + }, + { + "type": "title", + "bbox": [ + 278, + 186, + 333, + 198 + ], + "lines": [ + { + "bbox": [ + 276, + 185, + 336, + 200 + ], + "spans": [ + { + "bbox": [ + 276, + 185, + 336, + 200 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 143, + 210, + 468, + 375 + ], + "lines": [ + { + "bbox": [ + 141, + 210, + 469, + 222 + ], + "spans": [ + { + "bbox": [ + 141, + 210, + 469, + 222 + ], + "score": 1.0, + "content": "Domain adaptation for visual recognition has undergone great progress in the past", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 221, + 470, + 233 + ], + "spans": [ + { + "bbox": [ + 141, + 221, + 470, + 233 + ], + "score": 1.0, + "content": "few years. Nevertheless, most existing methods work in the so-called closed-set", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 232, + 470, + 245 + ], + "spans": [ + { + "bbox": [ + 141, + 232, + 470, + 245 + ], + "score": 1.0, + "content": "scenario, assuming that the classes depicted by the target images are exactly the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 140, + 241, + 469, + 257 + ], + "spans": [ + { + "bbox": [ + 140, + 241, + 469, + 257 + ], + "score": 1.0, + "content": "same as those of the source domain. In this paper, we tackle the more challeng-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 255, + 469, + 266 + ], + "spans": [ + { + "bbox": [ + 141, + 255, + 469, + 266 + ], + "score": 1.0, + "content": "ing, yet more realistic case of open-set domain adaptation, where new, unknown", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 265, + 469, + 278 + ], + "spans": [ + { + "bbox": [ + 141, + 265, + 469, + 278 + ], + "score": 1.0, + "content": "classes can be present in the target data. While, in the unsupervised scenario,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 276, + 469, + 288 + ], + "spans": [ + { + "bbox": [ + 141, + 276, + 469, + 288 + ], + "score": 1.0, + "content": "one cannot expect to be able to identify each specific new class, we aim to au-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 142, + 288, + 470, + 299 + ], + "spans": [ + { + "bbox": [ + 142, + 288, + 470, + 299 + ], + "score": 1.0, + "content": "tomatically detect which samples belong to these new classes and discard them", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 298, + 470, + 311 + ], + "spans": [ + { + "bbox": [ + 141, + 298, + 470, + 311 + ], + "score": 1.0, + "content": "from the recognition process. To this end, we rely on the intuition that the source", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 309, + 470, + 321 + ], + "spans": [ + { + "bbox": [ + 141, + 309, + 470, + 321 + ], + "score": 1.0, + "content": "and target samples depicting the known classes can be generated by a shared sub-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 321, + 469, + 331 + ], + "spans": [ + { + "bbox": [ + 141, + 321, + 469, + 331 + ], + "score": 1.0, + "content": "space, whereas the target samples from unknown classes come from a different,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 331, + 469, + 343 + ], + "spans": [ + { + "bbox": [ + 141, + 331, + 469, + 343 + ], + "score": 1.0, + "content": "private subspace. We therefore introduce a framework that factorizes the data into", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 342, + 469, + 353 + ], + "spans": [ + { + "bbox": [ + 141, + 342, + 469, + 353 + ], + "score": 1.0, + "content": "shared and private parts, while encouraging the shared representation to be dis-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 352, + 469, + 365 + ], + "spans": [ + { + "bbox": [ + 141, + 352, + 469, + 365 + ], + "score": 1.0, + "content": "criminative. Our experiments on standard benchmarks evidence that our approach", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 142, + 364, + 393, + 376 + ], + "spans": [ + { + "bbox": [ + 142, + 364, + 393, + 376 + ], + "score": 1.0, + "content": "outperforms the state of the art in open-set domain adaptation.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 16, + "bbox_fs": [ + 140, + 210, + 470, + 376 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 394, + 206, + 407 + ], + "lines": [ + { + "bbox": [ + 105, + 393, + 208, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 208, + 410 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 419, + 505, + 507 + ], + "lines": [ + { + "bbox": [ + 105, + 418, + 505, + 432 + ], + "spans": [ + { + "bbox": [ + 105, + 418, + 505, + 432 + ], + "score": 1.0, + "content": "In many practical machine learning scenarios, the test samples are drawn from a different distribu-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 429, + 505, + 443 + ], + "spans": [ + { + "bbox": [ + 105, + 429, + 505, + 443 + ], + "score": 1.0, + "content": "tion from the training ones, due to varying acquisition conditions, such as different data sources,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 439, + 505, + 454 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 505, + 454 + ], + "score": 1.0, + "content": "illumination conditions and cameras, in the context of visual recognition. Over the years, great", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 451, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 505, + 466 + ], + "score": 1.0, + "content": "progress has been achieved to tackle this problem, known has the domain shift. In particular, many", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 463, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 463, + 505, + 475 + ], + "score": 1.0, + "content": "methods aim to align the source (i.e., training) and target (i.e., test) distributions by learning domain-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 474, + 505, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 505, + 486 + ], + "score": 1.0, + "content": "invariant embeddings (Pan et al., 2011; Gong et al., 2012; Fernando et al., 2013; Sun et al., 2016),", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 485, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 505, + 497 + ], + "score": 1.0, + "content": "the most recent approaches relying on deep networks (Ganin & Lempitsky, 2014; Long et al., 2015;", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 496, + 426, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 426, + 508 + ], + "score": 1.0, + "content": "Bousmalis et al., 2016; Tzeng et al., 2017; Long et al., 2016a; Yan et al., 2017).", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 28.5, + "bbox_fs": [ + 105, + 418, + 505, + 508 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 513, + 505, + 677 + ], + "lines": [ + { + "bbox": [ + 106, + 512, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 106, + 512, + 505, + 525 + ], + "score": 1.0, + "content": "While effective, these methods work under the assumption that the source and target data contain ex-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 524, + 505, + 536 + ], + "spans": [ + { + "bbox": [ + 106, + 524, + 505, + 536 + ], + "score": 1.0, + "content": "actly the same classes. In practice, however, this assumption may easily be violated, as the target data", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 535, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 535, + 505, + 547 + ], + "score": 1.0, + "content": "will often contain additional classes that were not present within the source data. For example, when", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 545, + 505, + 558 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 505, + 558 + ], + "score": 1.0, + "content": "training a model to recognize office objects from images, as with the popular Office dataset (Saenko", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 104, + 554, + 505, + 570 + ], + "spans": [ + { + "bbox": [ + 104, + 554, + 505, + 570 + ], + "score": 1.0, + "content": "et al., 2010), one should still expect to see new objects, unobserved during training, when deploying", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 567, + 506, + 580 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 506, + 580 + ], + "score": 1.0, + "content": "the model in the real world. While one should not expect the model to recognize the specific class", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 578, + 506, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 506, + 591 + ], + "score": 1.0, + "content": "of such objects, at least in unsupervised domain adaptation where no target labels are provided, it", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 589, + 505, + 602 + ], + "spans": [ + { + "bbox": [ + 105, + 589, + 505, + 602 + ], + "score": 1.0, + "content": "would nonetheless be beneficial to identify these objects as unknown instead of misclassifying them.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 600, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 505, + 613 + ], + "score": 1.0, + "content": "This was the task addressed by Busto & Gall (2017) in their so-called open-set domain adaptation", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 611, + 506, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 611, + 506, + 624 + ], + "score": 1.0, + "content": "approach. This method aims to learn a mapping from the source samples to a subset of the target", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 104, + 621, + 506, + 636 + ], + "spans": [ + { + "bbox": [ + 104, + 621, + 506, + 636 + ], + "score": 1.0, + "content": "ones corresponding to those identified as coming from known classes. While reasonably effective,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 633, + 506, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 506, + 645 + ], + "score": 1.0, + "content": "this procedure involves alternatively solving for the mapping and the assignment of the samples to", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "score": 1.0, + "content": "known/unknown classes, which, as shown in our experiments, can be costly. Recently, Saito et al.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "(2018) introduced a deep learning framework for open-set domain adaptation, relying on adversarial", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 667, + 430, + 678 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 430, + 678 + ], + "score": 1.0, + "content": "training to separate the samples from the known classes from the unknown ones.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 40, + "bbox_fs": [ + 104, + 512, + 506, + 678 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 683, + 502, + 705 + ], + "lines": [ + { + "bbox": [ + 105, + 682, + 506, + 696 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 506, + 696 + ], + "score": 1.0, + "content": "In this paper, we introduce a novel approach to open-set domain adaptation based on learning a", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 694, + 504, + 705 + ], + "spans": [ + { + "bbox": [ + 106, + 694, + 504, + 705 + ], + "score": 1.0, + "content": "factorized representation of the source and target data. In essence, we seek to model the samples", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 82, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 96 + ], + "score": 1.0, + "content": "from the known classes with a low-dimensional subspace, shared by the source and target domains,", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "and the target samples from unknown classes with another subspace, specific to the target domain.", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 505, + 118 + ], + "score": 1.0, + "content": "We then make use of group sparsity to encourage each target sample to be reconstructed by only", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "score": 1.0, + "content": "one of these subspaces, which in turns lets us identify if this sample corresponds to a known or", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 505, + 140 + ], + "score": 1.0, + "content": "unknown class. We further show that we can obtain a more discriminative shared representation by", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 506, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 506, + 150 + ], + "score": 1.0, + "content": "jointly learning a linear classifier within our framework. Ultimately, our approach therefore allows", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 149, + 505, + 162 + ], + "spans": [ + { + "bbox": [ + 105, + 149, + 505, + 162 + ], + "score": 1.0, + "content": "us to jointly separate the target samples between known and unknown classes and represent the", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 159, + 505, + 172 + ], + "spans": [ + { + "bbox": [ + 105, + 159, + 505, + 172 + ], + "score": 1.0, + "content": "source and target samples within a consistent, shared latent space. Note that our approach is more", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 170, + 505, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 170, + 505, + 182 + ], + "score": 1.0, + "content": "intuitive than (Bousmalis et al., 2016) for the open-set DA scenario in the sense that we model each", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 181, + 505, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 181, + 505, + 194 + ], + "score": 1.0, + "content": "target sample as being generated by either the shared subspace or the private one, which is crucial", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 192, + 505, + 205 + ], + "spans": [ + { + "bbox": [ + 105, + 192, + 505, + 205 + ], + "score": 1.0, + "content": "to identify the target samples depicting unknown classes. By contrast, in (Bousmalis et al., 2016),", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 203, + 505, + 215 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 505, + 215 + ], + "score": 1.0, + "content": "each sample is encoded as a mixture of shared and private representations, which does not provide", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 214, + 346, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 214, + 346, + 226 + ], + "score": 1.0, + "content": "information to discriminate samples from unknown classes.", + "type": "text", + "cross_page": true + } + ], + "index": 12 + } + ], + "index": 48.5, + "bbox_fs": [ + 105, + 682, + 506, + 705 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 225 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 96 + ], + "score": 1.0, + "content": "from the known classes with a low-dimensional subspace, shared by the source and target domains,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "and the target samples from unknown classes with another subspace, specific to the target domain.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 505, + 118 + ], + "score": 1.0, + "content": "We then make use of group sparsity to encourage each target sample to be reconstructed by only", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "score": 1.0, + "content": "one of these subspaces, which in turns lets us identify if this sample corresponds to a known or", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 505, + 140 + ], + "score": 1.0, + "content": "unknown class. We further show that we can obtain a more discriminative shared representation by", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 506, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 506, + 150 + ], + "score": 1.0, + "content": "jointly learning a linear classifier within our framework. Ultimately, our approach therefore allows", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 149, + 505, + 162 + ], + "spans": [ + { + "bbox": [ + 105, + 149, + 505, + 162 + ], + "score": 1.0, + "content": "us to jointly separate the target samples between known and unknown classes and represent the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 159, + 505, + 172 + ], + "spans": [ + { + "bbox": [ + 105, + 159, + 505, + 172 + ], + "score": 1.0, + "content": "source and target samples within a consistent, shared latent space. Note that our approach is more", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 170, + 505, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 170, + 505, + 182 + ], + "score": 1.0, + "content": "intuitive than (Bousmalis et al., 2016) for the open-set DA scenario in the sense that we model each", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 181, + 505, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 181, + 505, + 194 + ], + "score": 1.0, + "content": "target sample as being generated by either the shared subspace or the private one, which is crucial", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 192, + 505, + 205 + ], + "spans": [ + { + "bbox": [ + 105, + 192, + 505, + 205 + ], + "score": 1.0, + "content": "to identify the target samples depicting unknown classes. By contrast, in (Bousmalis et al., 2016),", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 203, + 505, + 215 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 505, + 215 + ], + "score": 1.0, + "content": "each sample is encoded as a mixture of shared and private representations, which does not provide", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 214, + 346, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 214, + 346, + 226 + ], + "score": 1.0, + "content": "information to discriminate samples from unknown classes.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 107, + 231, + 505, + 297 + ], + "lines": [ + { + "bbox": [ + 106, + 230, + 504, + 243 + ], + "spans": [ + { + "bbox": [ + 106, + 230, + 504, + 243 + ], + "score": 1.0, + "content": "We demonstrate the effectiveness of our approach on several open-set domain adaptation bench-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 242, + 505, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 505, + 254 + ], + "score": 1.0, + "content": "marks for visual object recognition. Our method consistently and significantly outperforms the", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 252, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 505, + 266 + ], + "score": 1.0, + "content": "technique of Busto & Gall (2017) on all benchmarks, as well as the end-to-end learning approach", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 264, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 106, + 264, + 505, + 276 + ], + "score": 1.0, + "content": "of Saito et al. (2018) on the Office dataset, thus showing the benefits of learning shared and private", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 275, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 505, + 288 + ], + "score": 1.0, + "content": "representations corresponding to the known and unknown classes, respectively. Furthermore, it is", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 286, + 408, + 298 + ], + "spans": [ + { + "bbox": [ + 105, + 286, + 408, + 298 + ], + "score": 1.0, + "content": "faster than the algorithm of Busto & Gall (2017) by an order of magnitude.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 15.5 + }, + { + "type": "title", + "bbox": [ + 108, + 310, + 211, + 323 + ], + "lines": [ + { + "bbox": [ + 104, + 309, + 213, + 326 + ], + "spans": [ + { + "bbox": [ + 104, + 309, + 213, + 326 + ], + "score": 1.0, + "content": "2 RELATED WORK", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 330, + 505, + 407 + ], + "lines": [ + { + "bbox": [ + 106, + 329, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 505, + 343 + ], + "score": 1.0, + "content": "Domain adaptation for visual recognition has become increasingly popular over the past few years,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 341, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 341, + 505, + 353 + ], + "score": 1.0, + "content": "in large part thanks to the benchmark Office dataset of Saenko et al. (2010). A natural approach to", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 352, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 505, + 364 + ], + "score": 1.0, + "content": "tackling the domain shift consists of learning a transformation of the data such that the distributions", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 363, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 505, + 376 + ], + "score": 1.0, + "content": "of the source and target samples are as similar as possible in the resulting space (Baktashmotlagh", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 374, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 505, + 386 + ], + "score": 1.0, + "content": "et al., 2014; 2013; Sun et al., 2016). Instead of learning a transformation of the data, other methods", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 385, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 505, + 397 + ], + "score": 1.0, + "content": "have been proposed to re-weight the source samples, so as to rely more strongly on those that look", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 396, + 394, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 394, + 408 + ], + "score": 1.0, + "content": "similar to the target ones (Quiñonero C. et al., 2009; Gong et al., 2013).", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 107, + 412, + 505, + 511 + ], + "lines": [ + { + "bbox": [ + 105, + 412, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 505, + 425 + ], + "score": 1.0, + "content": "With the advent of deep learning for visual recognition, domain adaptation research also eventually", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 423, + 506, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 506, + 437 + ], + "score": 1.0, + "content": "turned to exploiting deep networks. While it was initially shown that deep features were more", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 434, + 505, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 505, + 448 + ], + "score": 1.0, + "content": "robust than handcrafted ones to the domain shift (Donahue et al., 2013), translating the above-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 444, + 506, + 460 + ], + "spans": [ + { + "bbox": [ + 104, + 444, + 506, + 460 + ], + "score": 1.0, + "content": "mentioned distribution-matching ideas to end-to-end learning proved even more effective (Tzeng", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 456, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 506, + 469 + ], + "score": 1.0, + "content": "et al., 2014; Long et al., 2015; 2016b; Rozantsev et al., 2018; Sun & Saenko, 2016). In this context,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 467, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 505, + 480 + ], + "score": 1.0, + "content": "other ideas were introduced, such as learning intermediate representations to interpolate between", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 479, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 106, + 479, + 505, + 491 + ], + "score": 1.0, + "content": "the source and target domains (Chopra S. & R., 2013; Tzeng et al., 2015), the use of adversarial", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "score": 1.0, + "content": "domain classifiers (Ganin & Lempitsky, 2014; Tzeng et al., 2017), and additional reconstruction", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 501, + 238, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 238, + 513 + ], + "score": 1.0, + "content": "loss terms (Ghifary et al., 2016).", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 107, + 517, + 505, + 638 + ], + "lines": [ + { + "bbox": [ + 105, + 518, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 505, + 530 + ], + "score": 1.0, + "content": "Despite achieving great progress to tackle the domain shift, all the aforementioned methods are", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 528, + 506, + 541 + ], + "spans": [ + { + "bbox": [ + 106, + 528, + 506, + 541 + ], + "score": 1.0, + "content": "designed for the closed-set scenario, where the source and target data depict the exact same set of", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 539, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 505, + 552 + ], + "score": 1.0, + "content": "classes. Inspired by recent advances in open-set recognition (Bendale & Boult, 2015; Scheirer et al.,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 550, + 505, + 562 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 505, + 562 + ], + "score": 1.0, + "content": "2014), the work of Busto & Gall (2017) constitutes the first attempt to address the more realistic", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 561, + 505, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 505, + 573 + ], + "score": 1.0, + "content": "case where the target data contains samples from new, unknown classes. To achieve this, Busto", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 571, + 506, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 571, + 506, + 586 + ], + "score": 1.0, + "content": "& Gall (2017) proposed to jointly learn the assignments of the target samples to known/unknown", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 582, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 505, + 596 + ], + "score": 1.0, + "content": "classes and a mapping from the source data to the target samples depicting known classes. The", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "resulting learning problem was solved by alternatively optimizing for the assignments and for the", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "score": 1.0, + "content": "mapping, which can be costly. Very recently, a deep learning approach was proposed for open-set", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 616, + 505, + 629 + ], + "spans": [ + { + "bbox": [ + 106, + 616, + 505, + 629 + ], + "score": 1.0, + "content": "domain adaptation (Saito et al., 2018), relying on adversarial training to separate the unknown target", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 627, + 231, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 231, + 640 + ], + "score": 1.0, + "content": "samples from the known ones.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 107, + 644, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 644, + 506, + 656 + ], + "spans": [ + { + "bbox": [ + 106, + 644, + 506, + 656 + ], + "score": 1.0, + "content": "Here, we introduce a new solution to the open-set domain adaptation problem, where we model the", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "source and target data with subspaces. Subspace-based representations have proven effective for", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 666, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 506, + 678 + ], + "score": 1.0, + "content": "domain adaptation (Gong et al., 2012; Gopalan et al., 2014; Fernando et al., 2013). Here, however,", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "we exploit them in a different manner, based on the intuition that source samples and target samples", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "from the known classes can be generated by a shared subspace, whereas target samples from un-", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "known classes come from a private subspace. While the notion of shared-private representations has", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 710, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 506, + 723 + ], + "score": 1.0, + "content": "been exploited in the past, e.g., for multiview learning (Jia et al., 2010) and for closed-set domain", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 106, + 721, + 506, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 506, + 733 + ], + "score": 1.0, + "content": "adaptation (Bousmalis et al., 2016), the resulting techniques all use them to encode each sample as", + "type": "text" + } + ], + "index": 54 + } + ], + "index": 50.5 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 225 + ], + "lines": [], + "index": 6, + "bbox_fs": [ + 105, + 82, + 506, + 226 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 231, + 505, + 297 + ], + "lines": [ + { + "bbox": [ + 106, + 230, + 504, + 243 + ], + "spans": [ + { + "bbox": [ + 106, + 230, + 504, + 243 + ], + "score": 1.0, + "content": "We demonstrate the effectiveness of our approach on several open-set domain adaptation bench-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 242, + 505, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 505, + 254 + ], + "score": 1.0, + "content": "marks for visual object recognition. Our method consistently and significantly outperforms the", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 252, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 505, + 266 + ], + "score": 1.0, + "content": "technique of Busto & Gall (2017) on all benchmarks, as well as the end-to-end learning approach", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 264, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 106, + 264, + 505, + 276 + ], + "score": 1.0, + "content": "of Saito et al. (2018) on the Office dataset, thus showing the benefits of learning shared and private", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 275, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 505, + 288 + ], + "score": 1.0, + "content": "representations corresponding to the known and unknown classes, respectively. Furthermore, it is", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 286, + 408, + 298 + ], + "spans": [ + { + "bbox": [ + 105, + 286, + 408, + 298 + ], + "score": 1.0, + "content": "faster than the algorithm of Busto & Gall (2017) by an order of magnitude.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 230, + 505, + 298 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 310, + 211, + 323 + ], + "lines": [ + { + "bbox": [ + 104, + 309, + 213, + 326 + ], + "spans": [ + { + "bbox": [ + 104, + 309, + 213, + 326 + ], + "score": 1.0, + "content": "2 RELATED WORK", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 330, + 505, + 407 + ], + "lines": [ + { + "bbox": [ + 106, + 329, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 505, + 343 + ], + "score": 1.0, + "content": "Domain adaptation for visual recognition has become increasingly popular over the past few years,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 341, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 341, + 505, + 353 + ], + "score": 1.0, + "content": "in large part thanks to the benchmark Office dataset of Saenko et al. (2010). A natural approach to", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 352, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 505, + 364 + ], + "score": 1.0, + "content": "tackling the domain shift consists of learning a transformation of the data such that the distributions", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 363, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 505, + 376 + ], + "score": 1.0, + "content": "of the source and target samples are as similar as possible in the resulting space (Baktashmotlagh", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 374, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 505, + 386 + ], + "score": 1.0, + "content": "et al., 2014; 2013; Sun et al., 2016). Instead of learning a transformation of the data, other methods", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 385, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 505, + 397 + ], + "score": 1.0, + "content": "have been proposed to re-weight the source samples, so as to rely more strongly on those that look", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 396, + 394, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 394, + 408 + ], + "score": 1.0, + "content": "similar to the target ones (Quiñonero C. et al., 2009; Gong et al., 2013).", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 329, + 505, + 408 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 412, + 505, + 511 + ], + "lines": [ + { + "bbox": [ + 105, + 412, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 505, + 425 + ], + "score": 1.0, + "content": "With the advent of deep learning for visual recognition, domain adaptation research also eventually", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 423, + 506, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 506, + 437 + ], + "score": 1.0, + "content": "turned to exploiting deep networks. While it was initially shown that deep features were more", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 434, + 505, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 505, + 448 + ], + "score": 1.0, + "content": "robust than handcrafted ones to the domain shift (Donahue et al., 2013), translating the above-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 444, + 506, + 460 + ], + "spans": [ + { + "bbox": [ + 104, + 444, + 506, + 460 + ], + "score": 1.0, + "content": "mentioned distribution-matching ideas to end-to-end learning proved even more effective (Tzeng", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 456, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 506, + 469 + ], + "score": 1.0, + "content": "et al., 2014; Long et al., 2015; 2016b; Rozantsev et al., 2018; Sun & Saenko, 2016). In this context,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 467, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 505, + 480 + ], + "score": 1.0, + "content": "other ideas were introduced, such as learning intermediate representations to interpolate between", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 479, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 106, + 479, + 505, + 491 + ], + "score": 1.0, + "content": "the source and target domains (Chopra S. & R., 2013; Tzeng et al., 2015), the use of adversarial", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "score": 1.0, + "content": "domain classifiers (Ganin & Lempitsky, 2014; Tzeng et al., 2017), and additional reconstruction", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 501, + 238, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 238, + 513 + ], + "score": 1.0, + "content": "loss terms (Ghifary et al., 2016).", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 31, + "bbox_fs": [ + 104, + 412, + 506, + 513 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 517, + 505, + 638 + ], + "lines": [ + { + "bbox": [ + 105, + 518, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 505, + 530 + ], + "score": 1.0, + "content": "Despite achieving great progress to tackle the domain shift, all the aforementioned methods are", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 528, + 506, + 541 + ], + "spans": [ + { + "bbox": [ + 106, + 528, + 506, + 541 + ], + "score": 1.0, + "content": "designed for the closed-set scenario, where the source and target data depict the exact same set of", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 539, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 505, + 552 + ], + "score": 1.0, + "content": "classes. Inspired by recent advances in open-set recognition (Bendale & Boult, 2015; Scheirer et al.,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 550, + 505, + 562 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 505, + 562 + ], + "score": 1.0, + "content": "2014), the work of Busto & Gall (2017) constitutes the first attempt to address the more realistic", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 561, + 505, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 505, + 573 + ], + "score": 1.0, + "content": "case where the target data contains samples from new, unknown classes. To achieve this, Busto", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 571, + 506, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 571, + 506, + 586 + ], + "score": 1.0, + "content": "& Gall (2017) proposed to jointly learn the assignments of the target samples to known/unknown", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 582, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 505, + 596 + ], + "score": 1.0, + "content": "classes and a mapping from the source data to the target samples depicting known classes. The", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "resulting learning problem was solved by alternatively optimizing for the assignments and for the", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "score": 1.0, + "content": "mapping, which can be costly. Very recently, a deep learning approach was proposed for open-set", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 616, + 505, + 629 + ], + "spans": [ + { + "bbox": [ + 106, + 616, + 505, + 629 + ], + "score": 1.0, + "content": "domain adaptation (Saito et al., 2018), relying on adversarial training to separate the unknown target", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 627, + 231, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 231, + 640 + ], + "score": 1.0, + "content": "samples from the known ones.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 41, + "bbox_fs": [ + 105, + 518, + 506, + 640 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 644, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 644, + 506, + 656 + ], + "spans": [ + { + "bbox": [ + 106, + 644, + 506, + 656 + ], + "score": 1.0, + "content": "Here, we introduce a new solution to the open-set domain adaptation problem, where we model the", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "source and target data with subspaces. Subspace-based representations have proven effective for", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 666, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 506, + 678 + ], + "score": 1.0, + "content": "domain adaptation (Gong et al., 2012; Gopalan et al., 2014; Fernando et al., 2013). Here, however,", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "we exploit them in a different manner, based on the intuition that source samples and target samples", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "from the known classes can be generated by a shared subspace, whereas target samples from un-", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "known classes come from a private subspace. While the notion of shared-private representations has", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 710, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 506, + 723 + ], + "score": 1.0, + "content": "been exploited in the past, e.g., for multiview learning (Jia et al., 2010) and for closed-set domain", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 106, + 721, + 506, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 506, + 733 + ], + "score": 1.0, + "content": "adaptation (Bousmalis et al., 2016), the resulting techniques all use them to encode each sample as", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "a mixture of shared and private information. By contrast, here, we aim to model each target sample", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "as being generated by either the shared subspace or the private one, which is crucial to identify the", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 279, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 279, + 118 + ], + "score": 1.0, + "content": "target samples depicting unknown classes.", + "type": "text", + "cross_page": true + } + ], + "index": 2 + } + ], + "index": 50.5, + "bbox_fs": [ + 105, + 644, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "a mixture of shared and private information. By contrast, here, we aim to model each target sample", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "as being generated by either the shared subspace or the private one, which is crucial to identify the", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 279, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 279, + 118 + ], + "score": 1.0, + "content": "target samples depicting unknown classes.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 107, + 121, + 505, + 165 + ], + "lines": [ + { + "bbox": [ + 105, + 121, + 505, + 134 + ], + "spans": [ + { + "bbox": [ + 105, + 121, + 505, + 134 + ], + "score": 1.0, + "content": "Our experiments evidence that our open-set domain adaptation approach, based on shared-private", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 131, + 506, + 146 + ], + "spans": [ + { + "bbox": [ + 104, + 131, + 506, + 146 + ], + "score": 1.0, + "content": "representations, is more effective than the one of Busto & Gall (2017), consistently outperforming", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 104, + 142, + 505, + 157 + ], + "spans": [ + { + "bbox": [ + 104, + 142, + 505, + 157 + ], + "score": 1.0, + "content": "it on several datasets, and also faster by an order of magnitude. We also outperform the recent deep", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 154, + 488, + 166 + ], + "spans": [ + { + "bbox": [ + 105, + 154, + 488, + 166 + ], + "score": 1.0, + "content": "learning open-set domain adaptation framework of Saito et al. (2018) on the Office benchmark.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + }, + { + "type": "title", + "bbox": [ + 108, + 183, + 209, + 196 + ], + "lines": [ + { + "bbox": [ + 104, + 182, + 211, + 198 + ], + "spans": [ + { + "bbox": [ + 104, + 182, + 211, + 198 + ], + "score": 1.0, + "content": "3 OUR APPROACH", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 106, + 209, + 505, + 298 + ], + "lines": [ + { + "bbox": [ + 105, + 209, + 505, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 505, + 222 + ], + "score": 1.0, + "content": "The key idea behind our formulation is to find low-dimensional representations of the data, factor-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 221, + 505, + 233 + ], + "spans": [ + { + "bbox": [ + 106, + 221, + 505, + 233 + ], + "score": 1.0, + "content": "ized into a subspace shared by the source samples and the target ones coming from known classes", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 232, + 504, + 244 + ], + "spans": [ + { + "bbox": [ + 106, + 232, + 504, + 244 + ], + "score": 1.0, + "content": "and another subspace specific to the target samples from unknown classes. Note that, when refer-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 243, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 243, + 505, + 255 + ], + "score": 1.0, + "content": "ring to target samples from known classes, we do not mean that these samples are labeled, but rather", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 254, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 505, + 266 + ], + "score": 1.0, + "content": "that they belong to the same set of classes as the source data. As a matter of fact, throughout the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 265, + 506, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 265, + 506, + 278 + ], + "score": 1.0, + "content": "paper, we focus on the unsupervised domain adaptation scenario, where no target annotations are", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 276, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 505, + 288 + ], + "score": 1.0, + "content": "provided. In the remainder of this section, we first introduce the optimization problem at the heart", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 287, + 407, + 299 + ], + "spans": [ + { + "bbox": [ + 106, + 287, + 407, + 299 + ], + "score": 1.0, + "content": "of our approach, and then discuss two extensions of this basic formulation.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 11.5 + }, + { + "type": "title", + "bbox": [ + 106, + 313, + 474, + 325 + ], + "lines": [ + { + "bbox": [ + 105, + 313, + 475, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 313, + 475, + 326 + ], + "score": 1.0, + "content": "3.1 FRODA: FACTORIZED REPRESENTATIONS FOR OPEN-SET DOMAIN ADAPTATION", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 333, + 505, + 412 + ], + "lines": [ + { + "bbox": [ + 104, + 332, + 506, + 349 + ], + "spans": [ + { + "bbox": [ + 104, + 332, + 133, + 349 + ], + "score": 1.0, + "content": "Given", + "type": "text" + }, + { + "bbox": [ + 133, + 337, + 145, + 345 + ], + "score": 0.83, + "content": "n _ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 332, + 295, + 349 + ], + "score": 1.0, + "content": "source samples, grouped in a matrix", + "type": "text" + }, + { + "bbox": [ + 295, + 334, + 352, + 346 + ], + "score": 0.93, + "content": "\\pmb { X } _ { s } \\in \\mathbb { R } ^ { D \\times n _ { s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 332, + 371, + 349 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 371, + 336, + 382, + 346 + ], + "score": 0.84, + "content": "n _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 382, + 332, + 506, + 349 + ], + "score": 1.0, + "content": "target samples represented by", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 107, + 342, + 507, + 360 + ], + "spans": [ + { + "bbox": [ + 107, + 345, + 164, + 357 + ], + "score": 0.92, + "content": "\\pmb { X } _ { t } \\in \\mathbb { R } ^ { D \\times n _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 342, + 507, + 360 + ], + "score": 1.0, + "content": ", our goal is to estimate a low-dimensional representation of each sample, such that", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 355, + 506, + 371 + ], + "spans": [ + { + "bbox": [ + 105, + 355, + 506, + 371 + ], + "score": 1.0, + "content": "the source and target samples coming from the same classes are generated by a shared subspace,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 367, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 505, + 380 + ], + "score": 1.0, + "content": "whereas the target data from new, unknown classes are generated by a different, specific subspace.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 104, + 376, + 505, + 392 + ], + "spans": [ + { + "bbox": [ + 104, + 376, + 171, + 392 + ], + "score": 1.0, + "content": "To this end, let", + "type": "text" + }, + { + "bbox": [ + 171, + 378, + 221, + 389 + ], + "score": 0.92, + "content": "V \\in \\mathbb { R } ^ { D \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 376, + 427, + 392 + ], + "score": 1.0, + "content": "be the matrix encoding the shared subspace, with", + "type": "text" + }, + { + "bbox": [ + 428, + 379, + 460, + 389 + ], + "score": 0.91, + "content": "d \\ll D", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 376, + 482, + 392 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 483, + 379, + 505, + 389 + ], + "score": 0.87, + "content": "\\boldsymbol { U } \\in", + "type": "inline_equation" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 387, + 506, + 405 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 132, + 399 + ], + "score": 0.9, + "content": "\\mathbb { R } ^ { D \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 387, + 506, + 405 + ], + "score": 1.0, + "content": "the one representing the private subspace. A naive approach to finding the low-dimensional", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 401, + 305, + 414 + ], + "spans": [ + { + "bbox": [ + 106, + 401, + 305, + 414 + ], + "score": 1.0, + "content": "representations of the data would involve solving", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20 + }, + { + "type": "interline_equation", + "bbox": [ + 211, + 418, + 398, + 438 + ], + "lines": [ + { + "bbox": [ + 211, + 418, + 398, + 438 + ], + "spans": [ + { + "bbox": [ + 211, + 418, + 398, + 438 + ], + "score": 0.91, + "content": "\\operatorname* { m i n } _ { U , T , V , S } \\quad \\| X _ { t } - B T \\| _ { F } ^ { 2 } + \\alpha \\| X _ { s } - V S \\| _ { F } ^ { 2 } \\ ,", + "type": "interline_equation", + "image_path": "ccf432ea8a26c6aba8fb50180e084bef065608157fc15cbad90443853971b9f0.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 211, + 418, + 398, + 438 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 446, + 505, + 492 + ], + "lines": [ + { + "bbox": [ + 105, + 445, + 506, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 133, + 460 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 450, + 141, + 457 + ], + "score": 0.77, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 445, + 306, + 460 + ], + "score": 1.0, + "content": "sets the relative influence of both terms,", + "type": "text" + }, + { + "bbox": [ + 306, + 446, + 400, + 459 + ], + "score": 0.94, + "content": "B = [ V , U ] \\in \\mathbb { R } ^ { D \\times 2 d }", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 445, + 421, + 460 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 421, + 447, + 431, + 457 + ], + "score": 0.82, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 445, + 449, + 460 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 450, + 448, + 458, + 457 + ], + "score": 0.83, + "content": "_ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 445, + 506, + 460 + ], + "score": 1.0, + "content": "encode the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 457, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 505, + 471 + ], + "score": 1.0, + "content": "low-dimensional representations of the target and source data, respectively. This simple formulation,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 470, + 505, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 505, + 482 + ], + "score": 1.0, + "content": "however, does not aim to separate the target samples belonging to known classes from the unknown", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 480, + 488, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 488, + 492 + ], + "score": 1.0, + "content": "ones, and thus will represent each target sample as a mixture of shared and private information.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 107, + 497, + 504, + 564 + ], + "lines": [ + { + "bbox": [ + 105, + 497, + 505, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 491, + 509 + ], + "score": 1.0, + "content": "Intuitively, we would rather like each target sample to be generated by either the shared subspace", + "type": "text" + }, + { + "bbox": [ + 491, + 497, + 501, + 507 + ], + "score": 0.76, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 497, + 505, + 509 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 508, + 505, + 521 + ], + "spans": [ + { + "bbox": [ + 106, + 508, + 178, + 521 + ], + "score": 1.0, + "content": "or the private one", + "type": "text" + }, + { + "bbox": [ + 178, + 509, + 189, + 518 + ], + "score": 0.71, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 508, + 505, + 521 + ], + "score": 1.0, + "content": ". To address this, we propose to make use of a group sparsity regularizer on the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 519, + 504, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 416, + 532 + ], + "score": 1.0, + "content": "coefficients of the target samples. Specifically, we split the coefficient vector", + "type": "text" + }, + { + "bbox": [ + 416, + 519, + 427, + 530 + ], + "score": 0.87, + "content": "\\mathbf { \\delta } _ { \\mathbf { \\mathcal { T } } _ { i } }", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 519, + 499, + 532 + ], + "score": 1.0, + "content": "for target sample", + "type": "text" + }, + { + "bbox": [ + 499, + 520, + 504, + 529 + ], + "score": 0.75, + "content": "i", + "type": "inline_equation" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 530, + 505, + 542 + ], + "spans": [ + { + "bbox": [ + 106, + 530, + 149, + 542 + ], + "score": 1.0, + "content": "into a part", + "type": "text" + }, + { + "bbox": [ + 150, + 530, + 163, + 542 + ], + "score": 0.89, + "content": "\\mathbf { \\mathscr { T } } _ { i } ^ { v }", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 530, + 366, + 542 + ], + "score": 1.0, + "content": "that corresponds to the shared subspace and a part", + "type": "text" + }, + { + "bbox": [ + 367, + 530, + 381, + 542 + ], + "score": 0.89, + "content": "\\mathbf { \\mathcal { T } } _ { i } ^ { u }", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 530, + 505, + 542 + ], + "score": 1.0, + "content": "that corresponds to the private", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 540, + 505, + 553 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 505, + 553 + ], + "score": 1.0, + "content": "one. We then encourage that either of these two parts goes to zero for each sample. To this end, we", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 551, + 271, + 564 + ], + "spans": [ + { + "bbox": [ + 106, + 551, + 271, + 564 + ], + "score": 1.0, + "content": "therefore write the optimization problem", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 31.5 + }, + { + "type": "interline_equation", + "bbox": [ + 142, + 570, + 430, + 640 + ], + "lines": [ + { + "bbox": [ + 142, + 570, + 430, + 640 + ], + "spans": [ + { + "bbox": [ + 142, + 570, + 430, + 640 + ], + "score": 0.94, + "content": "\\begin{array} { r l } { \\displaystyle \\underset { U , T , V , S } { \\operatorname* { m i n } } } & { \\| X _ { t } - B T \\| _ { F } ^ { 2 } + \\alpha \\| X _ { s } - V S \\| _ { F } ^ { 2 } + \\lambda _ { 1 } \\displaystyle \\sum _ { i = 1 } ^ { n _ { t } } \\left( \\| T _ { i } ^ { v } \\| + \\| T _ { i } ^ { u } \\| \\right) } \\\\ { s . t . } & { \\displaystyle \\displaystyle \\sum _ { j = 1 } ^ { d } \\| U _ { j } \\| ^ { 2 } \\leq 1 , \\displaystyle \\sum _ { j = 1 } ^ { d } \\| V _ { j } \\| ^ { 2 } \\leq 1 , } \\end{array}", + "type": "interline_equation", + "image_path": "4eee80a340c3f3f33eae0d8c1930637e0924fdfa84202e223380c9d35fe39e64.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 142, + 570, + 430, + 593.3333333333334 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 142, + 593.3333333333334, + 430, + 616.6666666666667 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 142, + 616.6666666666667, + 430, + 640.0000000000001 + ], + "spans": [], + "index": 37 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 645, + 505, + 702 + ], + "lines": [ + { + "bbox": [ + 106, + 646, + 505, + 658 + ], + "spans": [ + { + "bbox": [ + 106, + 646, + 505, + 658 + ], + "score": 1.0, + "content": "where the constraints prevent the basis vectors of the subspaces from growing while the coefficients", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 656, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 106, + 656, + 264, + 669 + ], + "score": 1.0, + "content": "decrease (Lee et al., 2007), and where", + "type": "text" + }, + { + "bbox": [ + 264, + 657, + 276, + 668 + ], + "score": 0.88, + "content": "\\lambda _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 656, + 505, + 669 + ], + "score": 1.0, + "content": "is a scalar controlling the strength of the group sparsity", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 668, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 506, + 681 + ], + "score": 1.0, + "content": "regularizer. In essence, this formulation allows each target sample to be reconstructed from either", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 678, + 506, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 506, + 692 + ], + "score": 1.0, + "content": "the shared subspace or the private one, which reduces the influence of the samples from unknown", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 690, + 288, + 702 + ], + "spans": [ + { + "bbox": [ + 106, + 690, + 288, + 702 + ], + "score": 1.0, + "content": "classes on learning the shared representation.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 503, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Optimization. To solve equation 2 efficiently, we alternatively update one variable at a time while", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 720, + 390, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 390, + 733 + ], + "score": 1.0, + "content": "keeping the other ones fixed. Below, we describe the different updates.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 504, + 116 + ], + "lines": [], + "index": 1, + "bbox_fs": [ + 105, + 82, + 505, + 118 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 121, + 505, + 165 + ], + "lines": [ + { + "bbox": [ + 105, + 121, + 505, + 134 + ], + "spans": [ + { + "bbox": [ + 105, + 121, + 505, + 134 + ], + "score": 1.0, + "content": "Our experiments evidence that our open-set domain adaptation approach, based on shared-private", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 131, + 506, + 146 + ], + "spans": [ + { + "bbox": [ + 104, + 131, + 506, + 146 + ], + "score": 1.0, + "content": "representations, is more effective than the one of Busto & Gall (2017), consistently outperforming", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 104, + 142, + 505, + 157 + ], + "spans": [ + { + "bbox": [ + 104, + 142, + 505, + 157 + ], + "score": 1.0, + "content": "it on several datasets, and also faster by an order of magnitude. We also outperform the recent deep", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 154, + 488, + 166 + ], + "spans": [ + { + "bbox": [ + 105, + 154, + 488, + 166 + ], + "score": 1.0, + "content": "learning open-set domain adaptation framework of Saito et al. (2018) on the Office benchmark.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5, + "bbox_fs": [ + 104, + 121, + 506, + 166 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 183, + 209, + 196 + ], + "lines": [ + { + "bbox": [ + 104, + 182, + 211, + 198 + ], + "spans": [ + { + "bbox": [ + 104, + 182, + 211, + 198 + ], + "score": 1.0, + "content": "3 OUR APPROACH", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 106, + 209, + 505, + 298 + ], + "lines": [ + { + "bbox": [ + 105, + 209, + 505, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 505, + 222 + ], + "score": 1.0, + "content": "The key idea behind our formulation is to find low-dimensional representations of the data, factor-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 221, + 505, + 233 + ], + "spans": [ + { + "bbox": [ + 106, + 221, + 505, + 233 + ], + "score": 1.0, + "content": "ized into a subspace shared by the source samples and the target ones coming from known classes", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 232, + 504, + 244 + ], + "spans": [ + { + "bbox": [ + 106, + 232, + 504, + 244 + ], + "score": 1.0, + "content": "and another subspace specific to the target samples from unknown classes. Note that, when refer-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 243, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 243, + 505, + 255 + ], + "score": 1.0, + "content": "ring to target samples from known classes, we do not mean that these samples are labeled, but rather", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 254, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 505, + 266 + ], + "score": 1.0, + "content": "that they belong to the same set of classes as the source data. As a matter of fact, throughout the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 265, + 506, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 265, + 506, + 278 + ], + "score": 1.0, + "content": "paper, we focus on the unsupervised domain adaptation scenario, where no target annotations are", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 276, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 505, + 288 + ], + "score": 1.0, + "content": "provided. In the remainder of this section, we first introduce the optimization problem at the heart", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 287, + 407, + 299 + ], + "spans": [ + { + "bbox": [ + 106, + 287, + 407, + 299 + ], + "score": 1.0, + "content": "of our approach, and then discuss two extensions of this basic formulation.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 209, + 506, + 299 + ] + }, + { + "type": "title", + "bbox": [ + 106, + 313, + 474, + 325 + ], + "lines": [ + { + "bbox": [ + 105, + 313, + 475, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 313, + 475, + 326 + ], + "score": 1.0, + "content": "3.1 FRODA: FACTORIZED REPRESENTATIONS FOR OPEN-SET DOMAIN ADAPTATION", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 333, + 505, + 412 + ], + "lines": [ + { + "bbox": [ + 104, + 332, + 506, + 349 + ], + "spans": [ + { + "bbox": [ + 104, + 332, + 133, + 349 + ], + "score": 1.0, + "content": "Given", + "type": "text" + }, + { + "bbox": [ + 133, + 337, + 145, + 345 + ], + "score": 0.83, + "content": "n _ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 332, + 295, + 349 + ], + "score": 1.0, + "content": "source samples, grouped in a matrix", + "type": "text" + }, + { + "bbox": [ + 295, + 334, + 352, + 346 + ], + "score": 0.93, + "content": "\\pmb { X } _ { s } \\in \\mathbb { R } ^ { D \\times n _ { s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 332, + 371, + 349 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 371, + 336, + 382, + 346 + ], + "score": 0.84, + "content": "n _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 382, + 332, + 506, + 349 + ], + "score": 1.0, + "content": "target samples represented by", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 107, + 342, + 507, + 360 + ], + "spans": [ + { + "bbox": [ + 107, + 345, + 164, + 357 + ], + "score": 0.92, + "content": "\\pmb { X } _ { t } \\in \\mathbb { R } ^ { D \\times n _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 342, + 507, + 360 + ], + "score": 1.0, + "content": ", our goal is to estimate a low-dimensional representation of each sample, such that", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 355, + 506, + 371 + ], + "spans": [ + { + "bbox": [ + 105, + 355, + 506, + 371 + ], + "score": 1.0, + "content": "the source and target samples coming from the same classes are generated by a shared subspace,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 367, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 505, + 380 + ], + "score": 1.0, + "content": "whereas the target data from new, unknown classes are generated by a different, specific subspace.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 104, + 376, + 505, + 392 + ], + "spans": [ + { + "bbox": [ + 104, + 376, + 171, + 392 + ], + "score": 1.0, + "content": "To this end, let", + "type": "text" + }, + { + "bbox": [ + 171, + 378, + 221, + 389 + ], + "score": 0.92, + "content": "V \\in \\mathbb { R } ^ { D \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 376, + 427, + 392 + ], + "score": 1.0, + "content": "be the matrix encoding the shared subspace, with", + "type": "text" + }, + { + "bbox": [ + 428, + 379, + 460, + 389 + ], + "score": 0.91, + "content": "d \\ll D", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 376, + 482, + 392 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 483, + 379, + 505, + 389 + ], + "score": 0.87, + "content": "\\boldsymbol { U } \\in", + "type": "inline_equation" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 387, + 506, + 405 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 132, + 399 + ], + "score": 0.9, + "content": "\\mathbb { R } ^ { D \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 387, + 506, + 405 + ], + "score": 1.0, + "content": "the one representing the private subspace. A naive approach to finding the low-dimensional", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 401, + 305, + 414 + ], + "spans": [ + { + "bbox": [ + 106, + 401, + 305, + 414 + ], + "score": 1.0, + "content": "representations of the data would involve solving", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20, + "bbox_fs": [ + 104, + 332, + 507, + 414 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 211, + 418, + 398, + 438 + ], + "lines": [ + { + "bbox": [ + 211, + 418, + 398, + 438 + ], + "spans": [ + { + "bbox": [ + 211, + 418, + 398, + 438 + ], + "score": 0.91, + "content": "\\operatorname* { m i n } _ { U , T , V , S } \\quad \\| X _ { t } - B T \\| _ { F } ^ { 2 } + \\alpha \\| X _ { s } - V S \\| _ { F } ^ { 2 } \\ ,", + "type": "interline_equation", + "image_path": "ccf432ea8a26c6aba8fb50180e084bef065608157fc15cbad90443853971b9f0.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 211, + 418, + 398, + 438 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 446, + 505, + 492 + ], + "lines": [ + { + "bbox": [ + 105, + 445, + 506, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 133, + 460 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 450, + 141, + 457 + ], + "score": 0.77, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 445, + 306, + 460 + ], + "score": 1.0, + "content": "sets the relative influence of both terms,", + "type": "text" + }, + { + "bbox": [ + 306, + 446, + 400, + 459 + ], + "score": 0.94, + "content": "B = [ V , U ] \\in \\mathbb { R } ^ { D \\times 2 d }", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 445, + 421, + 460 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 421, + 447, + 431, + 457 + ], + "score": 0.82, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 445, + 449, + 460 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 450, + 448, + 458, + 457 + ], + "score": 0.83, + "content": "_ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 445, + 506, + 460 + ], + "score": 1.0, + "content": "encode the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 457, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 505, + 471 + ], + "score": 1.0, + "content": "low-dimensional representations of the target and source data, respectively. This simple formulation,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 470, + 505, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 505, + 482 + ], + "score": 1.0, + "content": "however, does not aim to separate the target samples belonging to known classes from the unknown", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 480, + 488, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 488, + 492 + ], + "score": 1.0, + "content": "ones, and thus will represent each target sample as a mixture of shared and private information.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 445, + 506, + 492 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 497, + 504, + 564 + ], + "lines": [ + { + "bbox": [ + 105, + 497, + 505, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 491, + 509 + ], + "score": 1.0, + "content": "Intuitively, we would rather like each target sample to be generated by either the shared subspace", + "type": "text" + }, + { + "bbox": [ + 491, + 497, + 501, + 507 + ], + "score": 0.76, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 497, + 505, + 509 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 508, + 505, + 521 + ], + "spans": [ + { + "bbox": [ + 106, + 508, + 178, + 521 + ], + "score": 1.0, + "content": "or the private one", + "type": "text" + }, + { + "bbox": [ + 178, + 509, + 189, + 518 + ], + "score": 0.71, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 508, + 505, + 521 + ], + "score": 1.0, + "content": ". To address this, we propose to make use of a group sparsity regularizer on the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 519, + 504, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 416, + 532 + ], + "score": 1.0, + "content": "coefficients of the target samples. Specifically, we split the coefficient vector", + "type": "text" + }, + { + "bbox": [ + 416, + 519, + 427, + 530 + ], + "score": 0.87, + "content": "\\mathbf { \\delta } _ { \\mathbf { \\mathcal { T } } _ { i } }", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 519, + 499, + 532 + ], + "score": 1.0, + "content": "for target sample", + "type": "text" + }, + { + "bbox": [ + 499, + 520, + 504, + 529 + ], + "score": 0.75, + "content": "i", + "type": "inline_equation" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 530, + 505, + 542 + ], + "spans": [ + { + "bbox": [ + 106, + 530, + 149, + 542 + ], + "score": 1.0, + "content": "into a part", + "type": "text" + }, + { + "bbox": [ + 150, + 530, + 163, + 542 + ], + "score": 0.89, + "content": "\\mathbf { \\mathscr { T } } _ { i } ^ { v }", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 530, + 366, + 542 + ], + "score": 1.0, + "content": "that corresponds to the shared subspace and a part", + "type": "text" + }, + { + "bbox": [ + 367, + 530, + 381, + 542 + ], + "score": 0.89, + "content": "\\mathbf { \\mathcal { T } } _ { i } ^ { u }", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 530, + 505, + 542 + ], + "score": 1.0, + "content": "that corresponds to the private", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 540, + 505, + 553 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 505, + 553 + ], + "score": 1.0, + "content": "one. We then encourage that either of these two parts goes to zero for each sample. To this end, we", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 551, + 271, + 564 + ], + "spans": [ + { + "bbox": [ + 106, + 551, + 271, + 564 + ], + "score": 1.0, + "content": "therefore write the optimization problem", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 31.5, + "bbox_fs": [ + 105, + 497, + 505, + 564 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 142, + 570, + 430, + 640 + ], + "lines": [ + { + "bbox": [ + 142, + 570, + 430, + 640 + ], + "spans": [ + { + "bbox": [ + 142, + 570, + 430, + 640 + ], + "score": 0.94, + "content": "\\begin{array} { r l } { \\displaystyle \\underset { U , T , V , S } { \\operatorname* { m i n } } } & { \\| X _ { t } - B T \\| _ { F } ^ { 2 } + \\alpha \\| X _ { s } - V S \\| _ { F } ^ { 2 } + \\lambda _ { 1 } \\displaystyle \\sum _ { i = 1 } ^ { n _ { t } } \\left( \\| T _ { i } ^ { v } \\| + \\| T _ { i } ^ { u } \\| \\right) } \\\\ { s . t . } & { \\displaystyle \\displaystyle \\sum _ { j = 1 } ^ { d } \\| U _ { j } \\| ^ { 2 } \\leq 1 , \\displaystyle \\sum _ { j = 1 } ^ { d } \\| V _ { j } \\| ^ { 2 } \\leq 1 , } \\end{array}", + "type": "interline_equation", + "image_path": "4eee80a340c3f3f33eae0d8c1930637e0924fdfa84202e223380c9d35fe39e64.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 142, + 570, + 430, + 593.3333333333334 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 142, + 593.3333333333334, + 430, + 616.6666666666667 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 142, + 616.6666666666667, + 430, + 640.0000000000001 + ], + "spans": [], + "index": 37 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 645, + 505, + 702 + ], + "lines": [ + { + "bbox": [ + 106, + 646, + 505, + 658 + ], + "spans": [ + { + "bbox": [ + 106, + 646, + 505, + 658 + ], + "score": 1.0, + "content": "where the constraints prevent the basis vectors of the subspaces from growing while the coefficients", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 656, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 106, + 656, + 264, + 669 + ], + "score": 1.0, + "content": "decrease (Lee et al., 2007), and where", + "type": "text" + }, + { + "bbox": [ + 264, + 657, + 276, + 668 + ], + "score": 0.88, + "content": "\\lambda _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 656, + 505, + 669 + ], + "score": 1.0, + "content": "is a scalar controlling the strength of the group sparsity", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 668, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 506, + 681 + ], + "score": 1.0, + "content": "regularizer. In essence, this formulation allows each target sample to be reconstructed from either", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 678, + 506, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 506, + 692 + ], + "score": 1.0, + "content": "the shared subspace or the private one, which reduces the influence of the samples from unknown", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 690, + 288, + 702 + ], + "spans": [ + { + "bbox": [ + 106, + 690, + 288, + 702 + ], + "score": 1.0, + "content": "classes on learning the shared representation.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40, + "bbox_fs": [ + 105, + 646, + 506, + 702 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 503, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Optimization. To solve equation 2 efficiently, we alternatively update one variable at a time while", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 720, + 390, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 390, + 733 + ], + "score": 1.0, + "content": "keeping the other ones fixed. Below, we describe the different updates.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5, + "bbox_fs": [ + 106, + 709, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 106, + 82, + 448, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 450, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 450, + 96 + ], + "score": 1.0, + "content": "Algorithm 1 : FRODA: Factorized Representations for Open-set Domain Adaptation", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 107, + 101, + 135, + 111 + ], + "lines": [ + { + "bbox": [ + 107, + 101, + 135, + 111 + ], + "spans": [], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 123, + 111, + 302, + 145 + ], + "lines": [ + { + "bbox": [ + 124, + 108, + 263, + 124 + ], + "spans": [ + { + "bbox": [ + 124, + 110, + 180, + 122 + ], + "score": 0.86, + "content": "\\pmb { X } _ { s } \\in \\mathbb { R } ^ { D \\times n _ { s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 108, + 263, + 124 + ], + "score": 1.0, + "content": ": the source samples", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 124, + 120, + 259, + 135 + ], + "spans": [ + { + "bbox": [ + 124, + 122, + 178, + 133 + ], + "score": 0.85, + "content": "\\pmb { X } _ { t } \\in \\mathbb { R } ^ { D \\times n _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 120, + 259, + 135 + ], + "score": 1.0, + "content": ": the target samples", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 123, + 132, + 303, + 146 + ], + "spans": [ + { + "bbox": [ + 123, + 134, + 154, + 144 + ], + "score": 0.85, + "content": "d \\ll D", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 132, + 303, + 146 + ], + "score": 1.0, + "content": ": the dimensionality of the subspaces", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 106, + 147, + 226, + 170 + ], + "lines": [ + { + "bbox": [ + 104, + 144, + 145, + 164 + ], + "spans": [ + { + "bbox": [ + 104, + 144, + 145, + 164 + ], + "score": 1.0, + "content": "Output:", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 123, + 156, + 226, + 170 + ], + "spans": [ + { + "bbox": [ + 123, + 156, + 226, + 170 + ], + "score": 0.71, + "content": "\\pmb { \\zeta } \\in \\mathbb { R } ^ { d \\times n _ { s } } , \\pmb { T } \\in \\mathbb { R } ^ { 2 d \\times n _ { t } }", + "type": "inline_equation", + "image_path": "54e9b6361cedeccada8f1c30741246a865437f64d6db311c4eb123598dcfe012.jpg" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "title", + "bbox": [ + 110, + 180, + 153, + 190 + ], + "lines": [ + { + "bbox": [ + 109, + 180, + 154, + 194 + ], + "spans": [ + { + "bbox": [ + 109, + 180, + 154, + 194 + ], + "score": 1.0, + "content": "Initialize:", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 123, + 191, + 317, + 214 + ], + "lines": [ + { + "bbox": [ + 123, + 191, + 194, + 203 + ], + "spans": [ + { + "bbox": [ + 123, + 191, + 194, + 203 + ], + "score": 0.82, + "content": "V P C A ( X _ { s } )", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 123, + 201, + 317, + 214 + ], + "spans": [ + { + "bbox": [ + 123, + 203, + 187, + 214 + ], + "score": 0.85, + "content": "U \\gets N u l l ( V )", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 201, + 302, + 214 + ], + "score": 1.0, + "content": "(i.e., truncated null space of", + "type": "text" + }, + { + "bbox": [ + 303, + 203, + 312, + 212 + ], + "score": 0.64, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 201, + 317, + 214 + ], + "score": 1.0, + "content": ")", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 111, + 223, + 452, + 312 + ], + "lines": [ + { + "bbox": [ + 110, + 224, + 358, + 237 + ], + "spans": [ + { + "bbox": [ + 110, + 224, + 162, + 237 + ], + "score": 1.0, + "content": "1: Compute", + "type": "text" + }, + { + "bbox": [ + 162, + 225, + 172, + 234 + ], + "score": 0.81, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 224, + 358, + 237 + ], + "score": 1.0, + "content": "from equation 5 by proximal gradient descent", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 109, + 234, + 431, + 248 + ], + "spans": [ + { + "bbox": [ + 109, + 234, + 162, + 248 + ], + "score": 1.0, + "content": "2: Compute", + "type": "text" + }, + { + "bbox": [ + 163, + 235, + 171, + 245 + ], + "score": 0.8, + "content": "_ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 234, + 347, + 248 + ], + "score": 1.0, + "content": "by solving the linear least-squares problem", + "type": "text" + }, + { + "bbox": [ + 348, + 234, + 431, + 247 + ], + "score": 0.5, + "content": "\\mathrm { m i n } _ { S } \\| X _ { s } - V S \\| _ { F } ^ { 2 }", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 109, + 245, + 154, + 259 + ], + "spans": [ + { + "bbox": [ + 109, + 245, + 154, + 259 + ], + "score": 1.0, + "content": "3: repeat", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 111, + 256, + 452, + 269 + ], + "spans": [ + { + "bbox": [ + 111, + 257, + 121, + 268 + ], + "score": 1.0, + "content": "4:", + "type": "text" + }, + { + "bbox": [ + 132, + 256, + 172, + 269 + ], + "score": 1.0, + "content": "Compute", + "type": "text" + }, + { + "bbox": [ + 172, + 257, + 183, + 267 + ], + "score": 0.74, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 256, + 452, + 269 + ], + "score": 1.0, + "content": "from equation 3 by the Lagrange dual method of (Lee et al., 2007)", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 111, + 267, + 451, + 280 + ], + "spans": [ + { + "bbox": [ + 111, + 268, + 121, + 279 + ], + "score": 1.0, + "content": "5:", + "type": "text" + }, + { + "bbox": [ + 132, + 267, + 172, + 280 + ], + "score": 1.0, + "content": "Compute", + "type": "text" + }, + { + "bbox": [ + 173, + 268, + 183, + 278 + ], + "score": 0.74, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 267, + 451, + 280 + ], + "score": 1.0, + "content": "from equation 4 by the Lagrange dual method of (Lee et al., 2007)", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 111, + 279, + 368, + 291 + ], + "spans": [ + { + "bbox": [ + 111, + 280, + 121, + 290 + ], + "score": 1.0, + "content": "6:", + "type": "text" + }, + { + "bbox": [ + 133, + 279, + 172, + 291 + ], + "score": 1.0, + "content": "Compute", + "type": "text" + }, + { + "bbox": [ + 172, + 279, + 182, + 289 + ], + "score": 0.79, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 279, + 368, + 291 + ], + "score": 1.0, + "content": "from equation 5 by proximal gradient descent", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 111, + 288, + 441, + 303 + ], + "spans": [ + { + "bbox": [ + 111, + 290, + 121, + 301 + ], + "score": 1.0, + "content": "7:", + "type": "text" + }, + { + "bbox": [ + 132, + 288, + 172, + 303 + ], + "score": 1.0, + "content": "Compute", + "type": "text" + }, + { + "bbox": [ + 173, + 290, + 181, + 300 + ], + "score": 0.76, + "content": "_ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 288, + 356, + 303 + ], + "score": 1.0, + "content": "by solving the linear least-squares problem", + "type": "text" + }, + { + "bbox": [ + 356, + 289, + 441, + 302 + ], + "score": 0.79, + "content": "\\operatorname* { m i n } _ { S } \\| X _ { s } - V S \\| _ { F } ^ { 2 }", + "type": "inline_equation" + } + ], + "index": 16 + }, + { + "bbox": [ + 111, + 301, + 198, + 314 + ], + "spans": [ + { + "bbox": [ + 111, + 301, + 198, + 314 + ], + "score": 1.0, + "content": "8: until convergence", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 105, + 338, + 464, + 351 + ], + "lines": [ + { + "bbox": [ + 107, + 336, + 465, + 353 + ], + "spans": [ + { + "bbox": [ + 107, + 339, + 117, + 349 + ], + "score": 0.72, + "content": "\\textbf { { B } }", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 336, + 267, + 353 + ], + "score": 1.0, + "content": "-minimization: Given the coefficients", + "type": "text" + }, + { + "bbox": [ + 268, + 339, + 276, + 349 + ], + "score": 0.83, + "content": "_ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 336, + 294, + 353 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 294, + 339, + 303, + 349 + ], + "score": 0.8, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 336, + 387, + 353 + ], + "score": 1.0, + "content": ", we update the tuple", + "type": "text" + }, + { + "bbox": [ + 388, + 339, + 419, + 351 + ], + "score": 0.92, + "content": "( U , V )", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 336, + 465, + 353 + ], + "score": 1.0, + "content": "by solving", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "interline_equation", + "bbox": [ + 215, + 357, + 394, + 393 + ], + "lines": [ + { + "bbox": [ + 215, + 357, + 394, + 393 + ], + "spans": [ + { + "bbox": [ + 215, + 357, + 394, + 393 + ], + "score": 0.94, + "content": "\\operatorname* { m i n } _ { U } \\quad \\lVert A - U T ^ { u } \\rVert _ { F } ^ { 2 } \\quad s . t . \\sum _ { j = 1 } ^ { d } \\lVert U _ { j } \\rVert ^ { 2 } \\leq 1 ,", + "type": "interline_equation", + "image_path": "5a9a0aa985c90f4dd588532c6b7ffad64001c89dca70529fb6eafbbff051f83d.jpg" + } + ] + } + ], + "index": 19.5, + "virtual_lines": [ + { + "bbox": [ + 215, + 357, + 394, + 375.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 215, + 375.0, + 394, + 393.0 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 399, + 218, + 411 + ], + "lines": [ + { + "bbox": [ + 106, + 398, + 218, + 413 + ], + "spans": [ + { + "bbox": [ + 106, + 398, + 127, + 413 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 127, + 400, + 196, + 411 + ], + "score": 0.92, + "content": "A = X _ { t } - V T ^ { v }", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 398, + 218, + 413 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "interline_equation", + "bbox": [ + 174, + 417, + 434, + 453 + ], + "lines": [ + { + "bbox": [ + 174, + 417, + 434, + 453 + ], + "spans": [ + { + "bbox": [ + 174, + 417, + 434, + 453 + ], + "score": 0.93, + "content": "\\operatorname* { m i n } _ { \\pmb { V } } \\quad \\| \\pmb { A } ^ { \\prime } - \\pmb { V T } ^ { v } \\| _ { F } ^ { 2 } + \\alpha \\| \\pmb { X } _ { s } - \\pmb { V S } \\| _ { F } ^ { 2 } \\quad s . t . \\sum _ { j = 1 } ^ { d } \\| \\pmb { V } _ { j } \\| ^ { 2 } \\leq 1 ,", + "type": "interline_equation", + "image_path": "606708df43996c4c3ee16011056478d696446a214d52ff399281d4c293088779.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 174, + 417, + 434, + 429.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 174, + 429.0, + 434, + 441.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 174, + 441.0, + 434, + 453.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 460, + 504, + 483 + ], + "lines": [ + { + "bbox": [ + 106, + 459, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 106, + 459, + 127, + 473 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 127, + 460, + 200, + 472 + ], + "score": 0.91, + "content": "A ^ { \\prime } = X _ { t } - U T ^ { u } ", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 459, + 505, + 473 + ], + "score": 1.0, + "content": ". These two sub-problems can be solved efficiently using the Lagrange dual", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 470, + 503, + 485 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 503, + 485 + ], + "score": 1.0, + "content": "formulation introduced in (Lee et al., 2007) to update the basis in a standard sparse coding context.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 106, + 488, + 498, + 501 + ], + "lines": [ + { + "bbox": [ + 107, + 488, + 497, + 502 + ], + "spans": [ + { + "bbox": [ + 107, + 490, + 115, + 498 + ], + "score": 0.73, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 116, + 488, + 331, + 502 + ], + "score": 1.0, + "content": "-minimization: Minimizing equation 2 with respect to", + "type": "text" + }, + { + "bbox": [ + 332, + 489, + 341, + 498 + ], + "score": 0.81, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 488, + 497, + 502 + ], + "score": 1.0, + "content": ", with all other parameters fixed, yields", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "interline_equation", + "bbox": [ + 204, + 507, + 405, + 540 + ], + "lines": [ + { + "bbox": [ + 204, + 507, + 405, + 540 + ], + "spans": [ + { + "bbox": [ + 204, + 507, + 405, + 540 + ], + "score": 0.94, + "content": "\\operatorname* { m i n } _ { \\pmb { T } } \\| \\pmb { X } _ { t } - \\pmb { B } \\pmb { T } \\| _ { F } ^ { 2 } + \\lambda _ { 1 } \\sum _ { i = 1 } ^ { n _ { t } } ( \\| \\pmb { T } _ { i } ^ { v } \\| + \\| \\pmb { T } _ { i } ^ { u } \\| ) ~ ,", + "type": "interline_equation", + "image_path": "06955a667ea2ca0823275ef5c34c9e6f612d7e3b425b334e8688f43b6b265c89.jpg" + } + ] + } + ], + "index": 28.5, + "virtual_lines": [ + { + "bbox": [ + 204, + 507, + 405, + 523.5 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 204, + 523.5, + 405, + 540.0 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 546, + 504, + 580 + ], + "lines": [ + { + "bbox": [ + 106, + 546, + 506, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 506, + 558 + ], + "score": 1.0, + "content": "which can be solved efficiently via the proximal gradient method (Mairal et al., 2014). In other", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 558, + 505, + 570 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 271, + 570 + ], + "score": 1.0, + "content": "words, for each target sample, we obtain", + "type": "text" + }, + { + "bbox": [ + 271, + 558, + 280, + 567 + ], + "score": 0.82, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 558, + 505, + 570 + ], + "score": 1.0, + "content": "using group sparse coding to determine to which repre-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 568, + 331, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 331, + 581 + ], + "score": 1.0, + "content": "sentation, shared or private, each target sample belongs.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 106, + 585, + 503, + 607 + ], + "lines": [ + { + "bbox": [ + 107, + 584, + 505, + 598 + ], + "spans": [ + { + "bbox": [ + 107, + 586, + 115, + 595 + ], + "score": 0.73, + "content": "_ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 115, + 584, + 317, + 598 + ], + "score": 1.0, + "content": "-minimization: Solving equation 2 with respect to", + "type": "text" + }, + { + "bbox": [ + 317, + 586, + 325, + 595 + ], + "score": 0.8, + "content": "_ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 584, + 505, + 598 + ], + "score": 1.0, + "content": ", with all the other parameters fixed, reduces", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 595, + 378, + 608 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 378, + 608 + ], + "score": 1.0, + "content": "to a linear least-squares problem, which has a closed-form solution.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + }, + { + "type": "text", + "bbox": [ + 106, + 612, + 505, + 658 + ], + "lines": [ + { + "bbox": [ + 106, + 613, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 106, + 613, + 250, + 624 + ], + "score": 1.0, + "content": "To start optimization, we initialize", + "type": "text" + }, + { + "bbox": [ + 250, + 613, + 261, + 623 + ], + "score": 0.72, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 613, + 469, + 624 + ], + "score": 1.0, + "content": "as the PCA subspace of the source data, and take", + "type": "text" + }, + { + "bbox": [ + 469, + 613, + 479, + 623 + ], + "score": 0.76, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 613, + 505, + 624 + ], + "score": 1.0, + "content": "as its", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 624, + 506, + 636 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 343, + 636 + ], + "score": 1.0, + "content": "truncated null space. We then obtain the corresponding", + "type": "text" + }, + { + "bbox": [ + 343, + 624, + 353, + 634 + ], + "score": 0.76, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 624, + 374, + 636 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 374, + 624, + 383, + 634 + ], + "score": 0.8, + "content": "_ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 624, + 506, + 636 + ], + "score": 1.0, + "content": "as described above and start", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 634, + 505, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 634, + 505, + 647 + ], + "score": 1.0, + "content": "iterating. The pseudo-code of FRODA is provided in Algorithm 1. The steps are repeated until", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 646, + 503, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 646, + 503, + 658 + ], + "score": 1.0, + "content": "convergence, which typically occurs around 50 iterations, with each iteration taking roughly 0.05s.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 106, + 665, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 257, + 678 + ], + "score": 1.0, + "content": "Inference. After obtaining the final", + "type": "text" + }, + { + "bbox": [ + 257, + 666, + 268, + 676 + ], + "score": 0.81, + "content": "2 d", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 665, + 390, + 678 + ], + "score": 1.0, + "content": "-dimensional representation of", + "type": "text" + }, + { + "bbox": [ + 391, + 666, + 400, + 676 + ], + "score": 0.81, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "for the target samples, we", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 675, + 504, + 691 + ], + "spans": [ + { + "bbox": [ + 104, + 675, + 208, + 691 + ], + "score": 1.0, + "content": "determine if each sample", + "type": "text" + }, + { + "bbox": [ + 209, + 678, + 214, + 687 + ], + "score": 0.7, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 675, + 489, + 691 + ], + "score": 1.0, + "content": "belongs to known classes or unknown ones based on the coefficients", + "type": "text" + }, + { + "bbox": [ + 489, + 677, + 504, + 689 + ], + "score": 0.87, + "content": "\\mathbf { \\mathcal { T } } _ { i } ^ { u }", + "type": "inline_equation" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 686, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 104, + 686, + 123, + 701 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 688, + 137, + 699 + ], + "score": 0.88, + "content": "\\mathbf { \\mathit { T } } _ { i } ^ { v }", + "type": "inline_equation" + }, + { + "bbox": [ + 137, + 686, + 289, + 701 + ], + "score": 1.0, + "content": ". More specifically, given a threshold", + "type": "text" + }, + { + "bbox": [ + 289, + 690, + 295, + 698 + ], + "score": 0.72, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 686, + 505, + 701 + ], + "score": 1.0, + "content": ", a target sample is assigned to the unknown classes", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 116, + 712 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 117, + 699, + 188, + 711 + ], + "score": 0.91, + "content": "\\| \\mathbfcal { T } _ { i } ^ { \\dot { v } } \\| / \\| \\mathbfcal { T } _ { i } ^ { u } \\| \\leq \\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 698, + 505, + 712 + ], + "score": 1.0, + "content": ", which suggests that it can be well-reconstructed by the private subspace. In", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 173, + 722 + ], + "score": 1.0, + "content": "the presence of", + "type": "text" + }, + { + "bbox": [ + 173, + 711, + 182, + 720 + ], + "score": 0.78, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 709, + 316, + 722 + ], + "score": 1.0, + "content": "known classes, we then train a", + "type": "text" + }, + { + "bbox": [ + 317, + 710, + 352, + 722 + ], + "score": 0.9, + "content": "( C + 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 709, + 494, + 722 + ], + "score": 1.0, + "content": "-way classifier by augmenting the", + "type": "text" + }, + { + "bbox": [ + 494, + 710, + 501, + 720 + ], + "score": 0.71, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 718, + 488, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 718, + 205, + 733 + ], + "score": 1.0, + "content": "dimensional source data", + "type": "text" + }, + { + "bbox": [ + 206, + 721, + 214, + 731 + ], + "score": 0.79, + "content": "_ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 718, + 233, + 733 + ], + "score": 1.0, + "content": ", i.e.", + "type": "text" + }, + { + "bbox": [ + 233, + 720, + 280, + 731 + ], + "score": 0.92, + "content": "S \\in \\mathbb { R } ^ { d \\times n _ { s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 718, + 379, + 733 + ], + "score": 1.0, + "content": ", with the target samples", + "type": "text" + }, + { + "bbox": [ + 379, + 721, + 393, + 730 + ], + "score": 0.86, + "content": "\\mathbf { T } ^ { u }", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 718, + 488, + 733 + ], + "score": 1.0, + "content": "identified as unknown.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 41.5 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 12, + "width": 8 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 106, + 82, + 448, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 450, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 450, + 96 + ], + "score": 1.0, + "content": "Algorithm 1 : FRODA: Factorized Representations for Open-set Domain Adaptation", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 107, + 101, + 135, + 111 + ], + "lines": [ + { + "bbox": [ + 107, + 101, + 135, + 111 + ], + "spans": [], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 123, + 111, + 302, + 145 + ], + "lines": [ + { + "bbox": [ + 124, + 108, + 263, + 124 + ], + "spans": [ + { + "bbox": [ + 124, + 110, + 180, + 122 + ], + "score": 0.86, + "content": "\\pmb { X } _ { s } \\in \\mathbb { R } ^ { D \\times n _ { s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 108, + 263, + 124 + ], + "score": 1.0, + "content": ": the source samples", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 124, + 120, + 259, + 135 + ], + "spans": [ + { + "bbox": [ + 124, + 122, + 178, + 133 + ], + "score": 0.85, + "content": "\\pmb { X } _ { t } \\in \\mathbb { R } ^ { D \\times n _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 120, + 259, + 135 + ], + "score": 1.0, + "content": ": the target samples", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 123, + 132, + 303, + 146 + ], + "spans": [ + { + "bbox": [ + 123, + 134, + 154, + 144 + ], + "score": 0.85, + "content": "d \\ll D", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 132, + 303, + 146 + ], + "score": 1.0, + "content": ": the dimensionality of the subspaces", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3, + "bbox_fs": [ + 123, + 108, + 303, + 146 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 147, + 226, + 170 + ], + "lines": [ + { + "bbox": [ + 104, + 144, + 145, + 164 + ], + "spans": [ + { + "bbox": [ + 104, + 144, + 145, + 164 + ], + "score": 1.0, + "content": "Output:", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 123, + 156, + 226, + 170 + ], + "spans": [ + { + "bbox": [ + 123, + 156, + 226, + 170 + ], + "score": 0.71, + "content": "\\pmb { \\zeta } \\in \\mathbb { R } ^ { d \\times n _ { s } } , \\pmb { T } \\in \\mathbb { R } ^ { 2 d \\times n _ { t } }", + "type": "inline_equation", + "image_path": "54e9b6361cedeccada8f1c30741246a865437f64d6db311c4eb123598dcfe012.jpg" + } + ], + "index": 6 + } + ], + "index": 5.5, + "bbox_fs": [ + 104, + 144, + 226, + 170 + ] + }, + { + "type": "title", + "bbox": [ + 110, + 180, + 153, + 190 + ], + "lines": [ + { + "bbox": [ + 109, + 180, + 154, + 194 + ], + "spans": [ + { + "bbox": [ + 109, + 180, + 154, + 194 + ], + "score": 1.0, + "content": "Initialize:", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 123, + 191, + 317, + 214 + ], + "lines": [ + { + "bbox": [ + 123, + 191, + 194, + 203 + ], + "spans": [ + { + "bbox": [ + 123, + 191, + 194, + 203 + ], + "score": 0.82, + "content": "V P C A ( X _ { s } )", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 123, + 201, + 317, + 214 + ], + "spans": [ + { + "bbox": [ + 123, + 203, + 187, + 214 + ], + "score": 0.85, + "content": "U \\gets N u l l ( V )", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 201, + 302, + 214 + ], + "score": 1.0, + "content": "(i.e., truncated null space of", + "type": "text" + }, + { + "bbox": [ + 303, + 203, + 312, + 212 + ], + "score": 0.64, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 201, + 317, + 214 + ], + "score": 1.0, + "content": ")", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 123, + 191, + 317, + 214 + ] + }, + { + "type": "index", + "bbox": [ + 111, + 223, + 452, + 312 + ], + "lines": [ + { + "bbox": [ + 110, + 224, + 358, + 237 + ], + "spans": [ + { + "bbox": [ + 110, + 224, + 162, + 237 + ], + "score": 1.0, + "content": "1: Compute", + "type": "text" + }, + { + "bbox": [ + 162, + 225, + 172, + 234 + ], + "score": 0.81, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 224, + 358, + 237 + ], + "score": 1.0, + "content": "from equation 5 by proximal gradient descent", + "type": "text" + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 109, + 234, + 431, + 248 + ], + "spans": [ + { + "bbox": [ + 109, + 234, + 162, + 248 + ], + "score": 1.0, + "content": "2: Compute", + "type": "text" + }, + { + "bbox": [ + 163, + 235, + 171, + 245 + ], + "score": 0.8, + "content": "_ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 234, + 347, + 248 + ], + "score": 1.0, + "content": "by solving the linear least-squares problem", + "type": "text" + }, + { + "bbox": [ + 348, + 234, + 431, + 247 + ], + "score": 0.5, + "content": "\\mathrm { m i n } _ { S } \\| X _ { s } - V S \\| _ { F } ^ { 2 }", + "type": "inline_equation" + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 109, + 245, + 154, + 259 + ], + "spans": [ + { + "bbox": [ + 109, + 245, + 154, + 259 + ], + "score": 1.0, + "content": "3: repeat", + "type": "text" + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 111, + 256, + 452, + 269 + ], + "spans": [ + { + "bbox": [ + 111, + 257, + 121, + 268 + ], + "score": 1.0, + "content": "4:", + "type": "text" + }, + { + "bbox": [ + 132, + 256, + 172, + 269 + ], + "score": 1.0, + "content": "Compute", + "type": "text" + }, + { + "bbox": [ + 172, + 257, + 183, + 267 + ], + "score": 0.74, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 256, + 452, + 269 + ], + "score": 1.0, + "content": "from equation 3 by the Lagrange dual method of (Lee et al., 2007)", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 111, + 267, + 451, + 280 + ], + "spans": [ + { + "bbox": [ + 111, + 268, + 121, + 279 + ], + "score": 1.0, + "content": "5:", + "type": "text" + }, + { + "bbox": [ + 132, + 267, + 172, + 280 + ], + "score": 1.0, + "content": "Compute", + "type": "text" + }, + { + "bbox": [ + 173, + 268, + 183, + 278 + ], + "score": 0.74, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 267, + 451, + 280 + ], + "score": 1.0, + "content": "from equation 4 by the Lagrange dual method of (Lee et al., 2007)", + "type": "text" + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 111, + 279, + 368, + 291 + ], + "spans": [ + { + "bbox": [ + 111, + 280, + 121, + 290 + ], + "score": 1.0, + "content": "6:", + "type": "text" + }, + { + "bbox": [ + 133, + 279, + 172, + 291 + ], + "score": 1.0, + "content": "Compute", + "type": "text" + }, + { + "bbox": [ + 172, + 279, + 182, + 289 + ], + "score": 0.79, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 279, + 368, + 291 + ], + "score": 1.0, + "content": "from equation 5 by proximal gradient descent", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 111, + 288, + 441, + 303 + ], + "spans": [ + { + "bbox": [ + 111, + 290, + 121, + 301 + ], + "score": 1.0, + "content": "7:", + "type": "text" + }, + { + "bbox": [ + 132, + 288, + 172, + 303 + ], + "score": 1.0, + "content": "Compute", + "type": "text" + }, + { + "bbox": [ + 173, + 290, + 181, + 300 + ], + "score": 0.76, + "content": "_ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 288, + 356, + 303 + ], + "score": 1.0, + "content": "by solving the linear least-squares problem", + "type": "text" + }, + { + "bbox": [ + 356, + 289, + 441, + 302 + ], + "score": 0.79, + "content": "\\operatorname* { m i n } _ { S } \\| X _ { s } - V S \\| _ { F } ^ { 2 }", + "type": "inline_equation" + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 111, + 301, + 198, + 314 + ], + "spans": [ + { + "bbox": [ + 111, + 301, + 198, + 314 + ], + "score": 1.0, + "content": "8: until convergence", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true + } + ], + "index": 13.5, + "bbox_fs": [ + 109, + 224, + 452, + 314 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 338, + 464, + 351 + ], + "lines": [ + { + "bbox": [ + 107, + 336, + 465, + 353 + ], + "spans": [ + { + "bbox": [ + 107, + 339, + 117, + 349 + ], + "score": 0.72, + "content": "\\textbf { { B } }", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 336, + 267, + 353 + ], + "score": 1.0, + "content": "-minimization: Given the coefficients", + "type": "text" + }, + { + "bbox": [ + 268, + 339, + 276, + 349 + ], + "score": 0.83, + "content": "_ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 336, + 294, + 353 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 294, + 339, + 303, + 349 + ], + "score": 0.8, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 336, + 387, + 353 + ], + "score": 1.0, + "content": ", we update the tuple", + "type": "text" + }, + { + "bbox": [ + 388, + 339, + 419, + 351 + ], + "score": 0.92, + "content": "( U , V )", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 336, + 465, + 353 + ], + "score": 1.0, + "content": "by solving", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18, + "bbox_fs": [ + 107, + 336, + 465, + 353 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 215, + 357, + 394, + 393 + ], + "lines": [ + { + "bbox": [ + 215, + 357, + 394, + 393 + ], + "spans": [ + { + "bbox": [ + 215, + 357, + 394, + 393 + ], + "score": 0.94, + "content": "\\operatorname* { m i n } _ { U } \\quad \\lVert A - U T ^ { u } \\rVert _ { F } ^ { 2 } \\quad s . t . \\sum _ { j = 1 } ^ { d } \\lVert U _ { j } \\rVert ^ { 2 } \\leq 1 ,", + "type": "interline_equation", + "image_path": "5a9a0aa985c90f4dd588532c6b7ffad64001c89dca70529fb6eafbbff051f83d.jpg" + } + ] + } + ], + "index": 19.5, + "virtual_lines": [ + { + "bbox": [ + 215, + 357, + 394, + 375.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 215, + 375.0, + 394, + 393.0 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 399, + 218, + 411 + ], + "lines": [ + { + "bbox": [ + 106, + 398, + 218, + 413 + ], + "spans": [ + { + "bbox": [ + 106, + 398, + 127, + 413 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 127, + 400, + 196, + 411 + ], + "score": 0.92, + "content": "A = X _ { t } - V T ^ { v }", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 398, + 218, + 413 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21, + "bbox_fs": [ + 106, + 398, + 218, + 413 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 174, + 417, + 434, + 453 + ], + "lines": [ + { + "bbox": [ + 174, + 417, + 434, + 453 + ], + "spans": [ + { + "bbox": [ + 174, + 417, + 434, + 453 + ], + "score": 0.93, + "content": "\\operatorname* { m i n } _ { \\pmb { V } } \\quad \\| \\pmb { A } ^ { \\prime } - \\pmb { V T } ^ { v } \\| _ { F } ^ { 2 } + \\alpha \\| \\pmb { X } _ { s } - \\pmb { V S } \\| _ { F } ^ { 2 } \\quad s . t . \\sum _ { j = 1 } ^ { d } \\| \\pmb { V } _ { j } \\| ^ { 2 } \\leq 1 ,", + "type": "interline_equation", + "image_path": "606708df43996c4c3ee16011056478d696446a214d52ff399281d4c293088779.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 174, + 417, + 434, + 429.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 174, + 429.0, + 434, + 441.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 174, + 441.0, + 434, + 453.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 460, + 504, + 483 + ], + "lines": [ + { + "bbox": [ + 106, + 459, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 106, + 459, + 127, + 473 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 127, + 460, + 200, + 472 + ], + "score": 0.91, + "content": "A ^ { \\prime } = X _ { t } - U T ^ { u } ", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 459, + 505, + 473 + ], + "score": 1.0, + "content": ". These two sub-problems can be solved efficiently using the Lagrange dual", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 470, + 503, + 485 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 503, + 485 + ], + "score": 1.0, + "content": "formulation introduced in (Lee et al., 2007) to update the basis in a standard sparse coding context.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 459, + 505, + 485 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 488, + 498, + 501 + ], + "lines": [ + { + "bbox": [ + 107, + 488, + 497, + 502 + ], + "spans": [ + { + "bbox": [ + 107, + 490, + 115, + 498 + ], + "score": 0.73, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 116, + 488, + 331, + 502 + ], + "score": 1.0, + "content": "-minimization: Minimizing equation 2 with respect to", + "type": "text" + }, + { + "bbox": [ + 332, + 489, + 341, + 498 + ], + "score": 0.81, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 488, + 497, + 502 + ], + "score": 1.0, + "content": ", with all other parameters fixed, yields", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27, + "bbox_fs": [ + 107, + 488, + 497, + 502 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 204, + 507, + 405, + 540 + ], + "lines": [ + { + "bbox": [ + 204, + 507, + 405, + 540 + ], + "spans": [ + { + "bbox": [ + 204, + 507, + 405, + 540 + ], + "score": 0.94, + "content": "\\operatorname* { m i n } _ { \\pmb { T } } \\| \\pmb { X } _ { t } - \\pmb { B } \\pmb { T } \\| _ { F } ^ { 2 } + \\lambda _ { 1 } \\sum _ { i = 1 } ^ { n _ { t } } ( \\| \\pmb { T } _ { i } ^ { v } \\| + \\| \\pmb { T } _ { i } ^ { u } \\| ) ~ ,", + "type": "interline_equation", + "image_path": "06955a667ea2ca0823275ef5c34c9e6f612d7e3b425b334e8688f43b6b265c89.jpg" + } + ] + } + ], + "index": 28.5, + "virtual_lines": [ + { + "bbox": [ + 204, + 507, + 405, + 523.5 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 204, + 523.5, + 405, + 540.0 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 546, + 504, + 580 + ], + "lines": [ + { + "bbox": [ + 106, + 546, + 506, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 506, + 558 + ], + "score": 1.0, + "content": "which can be solved efficiently via the proximal gradient method (Mairal et al., 2014). In other", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 558, + 505, + 570 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 271, + 570 + ], + "score": 1.0, + "content": "words, for each target sample, we obtain", + "type": "text" + }, + { + "bbox": [ + 271, + 558, + 280, + 567 + ], + "score": 0.82, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 558, + 505, + 570 + ], + "score": 1.0, + "content": "using group sparse coding to determine to which repre-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 568, + 331, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 331, + 581 + ], + "score": 1.0, + "content": "sentation, shared or private, each target sample belongs.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 546, + 506, + 581 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 585, + 503, + 607 + ], + "lines": [ + { + "bbox": [ + 107, + 584, + 505, + 598 + ], + "spans": [ + { + "bbox": [ + 107, + 586, + 115, + 595 + ], + "score": 0.73, + "content": "_ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 115, + 584, + 317, + 598 + ], + "score": 1.0, + "content": "-minimization: Solving equation 2 with respect to", + "type": "text" + }, + { + "bbox": [ + 317, + 586, + 325, + 595 + ], + "score": 0.8, + "content": "_ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 584, + 505, + 598 + ], + "score": 1.0, + "content": ", with all the other parameters fixed, reduces", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 595, + 378, + 608 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 378, + 608 + ], + "score": 1.0, + "content": "to a linear least-squares problem, which has a closed-form solution.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5, + "bbox_fs": [ + 106, + 584, + 505, + 608 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 612, + 505, + 658 + ], + "lines": [ + { + "bbox": [ + 106, + 613, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 106, + 613, + 250, + 624 + ], + "score": 1.0, + "content": "To start optimization, we initialize", + "type": "text" + }, + { + "bbox": [ + 250, + 613, + 261, + 623 + ], + "score": 0.72, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 613, + 469, + 624 + ], + "score": 1.0, + "content": "as the PCA subspace of the source data, and take", + "type": "text" + }, + { + "bbox": [ + 469, + 613, + 479, + 623 + ], + "score": 0.76, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 613, + 505, + 624 + ], + "score": 1.0, + "content": "as its", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 624, + 506, + 636 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 343, + 636 + ], + "score": 1.0, + "content": "truncated null space. We then obtain the corresponding", + "type": "text" + }, + { + "bbox": [ + 343, + 624, + 353, + 634 + ], + "score": 0.76, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 624, + 374, + 636 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 374, + 624, + 383, + 634 + ], + "score": 0.8, + "content": "_ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 624, + 506, + 636 + ], + "score": 1.0, + "content": "as described above and start", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 634, + 505, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 634, + 505, + 647 + ], + "score": 1.0, + "content": "iterating. The pseudo-code of FRODA is provided in Algorithm 1. The steps are repeated until", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 646, + 503, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 646, + 503, + 658 + ], + "score": 1.0, + "content": "convergence, which typically occurs around 50 iterations, with each iteration taking roughly 0.05s.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36.5, + "bbox_fs": [ + 105, + 613, + 506, + 658 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 665, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 257, + 678 + ], + "score": 1.0, + "content": "Inference. After obtaining the final", + "type": "text" + }, + { + "bbox": [ + 257, + 666, + 268, + 676 + ], + "score": 0.81, + "content": "2 d", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 665, + 390, + 678 + ], + "score": 1.0, + "content": "-dimensional representation of", + "type": "text" + }, + { + "bbox": [ + 391, + 666, + 400, + 676 + ], + "score": 0.81, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "for the target samples, we", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 675, + 504, + 691 + ], + "spans": [ + { + "bbox": [ + 104, + 675, + 208, + 691 + ], + "score": 1.0, + "content": "determine if each sample", + "type": "text" + }, + { + "bbox": [ + 209, + 678, + 214, + 687 + ], + "score": 0.7, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 675, + 489, + 691 + ], + "score": 1.0, + "content": "belongs to known classes or unknown ones based on the coefficients", + "type": "text" + }, + { + "bbox": [ + 489, + 677, + 504, + 689 + ], + "score": 0.87, + "content": "\\mathbf { \\mathcal { T } } _ { i } ^ { u }", + "type": "inline_equation" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 686, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 104, + 686, + 123, + 701 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 688, + 137, + 699 + ], + "score": 0.88, + "content": "\\mathbf { \\mathit { T } } _ { i } ^ { v }", + "type": "inline_equation" + }, + { + "bbox": [ + 137, + 686, + 289, + 701 + ], + "score": 1.0, + "content": ". More specifically, given a threshold", + "type": "text" + }, + { + "bbox": [ + 289, + 690, + 295, + 698 + ], + "score": 0.72, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 686, + 505, + 701 + ], + "score": 1.0, + "content": ", a target sample is assigned to the unknown classes", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 116, + 712 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 117, + 699, + 188, + 711 + ], + "score": 0.91, + "content": "\\| \\mathbfcal { T } _ { i } ^ { \\dot { v } } \\| / \\| \\mathbfcal { T } _ { i } ^ { u } \\| \\leq \\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 698, + 505, + 712 + ], + "score": 1.0, + "content": ", which suggests that it can be well-reconstructed by the private subspace. In", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 173, + 722 + ], + "score": 1.0, + "content": "the presence of", + "type": "text" + }, + { + "bbox": [ + 173, + 711, + 182, + 720 + ], + "score": 0.78, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 709, + 316, + 722 + ], + "score": 1.0, + "content": "known classes, we then train a", + "type": "text" + }, + { + "bbox": [ + 317, + 710, + 352, + 722 + ], + "score": 0.9, + "content": "( C + 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 709, + 494, + 722 + ], + "score": 1.0, + "content": "-way classifier by augmenting the", + "type": "text" + }, + { + "bbox": [ + 494, + 710, + 501, + 720 + ], + "score": 0.71, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 718, + 488, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 718, + 205, + 733 + ], + "score": 1.0, + "content": "dimensional source data", + "type": "text" + }, + { + "bbox": [ + 206, + 721, + 214, + 731 + ], + "score": 0.79, + "content": "_ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 718, + 233, + 733 + ], + "score": 1.0, + "content": ", i.e.", + "type": "text" + }, + { + "bbox": [ + 233, + 720, + 280, + 731 + ], + "score": 0.92, + "content": "S \\in \\mathbb { R } ^ { d \\times n _ { s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 718, + 379, + 733 + ], + "score": 1.0, + "content": ", with the target samples", + "type": "text" + }, + { + "bbox": [ + 379, + 721, + 393, + 730 + ], + "score": 0.86, + "content": "\\mathbf { T } ^ { u }", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 718, + 488, + 733 + ], + "score": 1.0, + "content": "identified as unknown.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 41.5, + "bbox_fs": [ + 104, + 665, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 82, + 294, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 295, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 295, + 95 + ], + "score": 1.0, + "content": "3.2 D-FRODA: DISCRIMINATIVE FRODA", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 102, + 505, + 159 + ], + "lines": [ + { + "bbox": [ + 106, + 102, + 505, + 116 + ], + "spans": [ + { + "bbox": [ + 106, + 102, + 505, + 116 + ], + "score": 1.0, + "content": "The formulation above does not make use of the source labels at all during the representation learn-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 114, + 505, + 126 + ], + "spans": [ + { + "bbox": [ + 106, + 114, + 505, + 126 + ], + "score": 1.0, + "content": "ing stage. As such, it does not encourage the representation to be discriminative. To overcome this,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 123, + 505, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 123, + 505, + 139 + ], + "score": 1.0, + "content": "we extend our basic formulation to further account for the classification task at hand. Specifically,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 134, + 506, + 150 + ], + "spans": [ + { + "bbox": [ + 104, + 134, + 119, + 150 + ], + "score": 1.0, + "content": "let", + "type": "text" + }, + { + "bbox": [ + 119, + 135, + 224, + 148 + ], + "score": 0.92, + "content": "\\pmb { L } = [ l _ { 1 } \\dots l _ { n _ { s } } ] \\in \\mathbb { R } ^ { C \\times n _ { s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 134, + 425, + 150 + ], + "score": 1.0, + "content": "be the matrix containing the source labels, where", + "type": "text" + }, + { + "bbox": [ + 426, + 136, + 460, + 147 + ], + "score": 0.92, + "content": "\\boldsymbol { l } _ { i } \\in \\mathbb { R } ^ { C }", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 134, + 506, + 150 + ], + "score": 1.0, + "content": "represents", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 147, + 471, + 159 + ], + "spans": [ + { + "bbox": [ + 106, + 147, + 282, + 159 + ], + "score": 1.0, + "content": "the one-hot encoding of the label of sample", + "type": "text" + }, + { + "bbox": [ + 282, + 148, + 286, + 157 + ], + "score": 0.68, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 147, + 471, + 159 + ], + "score": 1.0, + "content": ". We then write our D-FRODA formulation as", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3 + }, + { + "type": "interline_equation", + "bbox": [ + 112, + 164, + 488, + 234 + ], + "lines": [ + { + "bbox": [ + 112, + 164, + 488, + 234 + ], + "spans": [ + { + "bbox": [ + 112, + 164, + 488, + 234 + ], + "score": 0.93, + "content": "\\begin{array} { r l } { \\displaystyle \\underset { U , T , V , S , W } { \\operatorname* { m i n } } } & { \\| X _ { t } - B T \\| _ { F } ^ { 2 } + \\alpha \\| X _ { s } - V S \\| _ { F } ^ { 2 } + \\beta \\| L - W S \\| _ { F } ^ { 2 } + \\lambda _ { 1 } \\displaystyle \\sum _ { i = 1 } ^ { n _ { t } } ( \\| T _ { i } ^ { v } \\| + \\| T _ { i } ^ { u } \\| ) ) } \\\\ { \\displaystyle s . t . } & { \\displaystyle \\displaystyle \\sum _ { j = 1 } ^ { d } \\| U _ { j } \\| ^ { 2 } \\leq 1 , \\displaystyle \\sum _ { j = 1 } ^ { d } \\| V _ { j } \\| ^ { 2 } \\leq 1 , } \\end{array}", + "type": "interline_equation", + "image_path": "a4cec1062f3cabe4da7a316f96ab8b73126525037273f676a43ae2a72aa32461.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 112, + 164, + 488, + 187.33333333333334 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 112, + 187.33333333333334, + 488, + 210.66666666666669 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 112, + 210.66666666666669, + 488, + 234.00000000000003 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 239, + 502, + 252 + ], + "lines": [ + { + "bbox": [ + 105, + 236, + 503, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 236, + 133, + 254 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 239, + 183, + 250 + ], + "score": 0.92, + "content": "W \\in \\mathbb { R } ^ { C \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 236, + 503, + 254 + ], + "score": 1.0, + "content": "is the matrix containing the parameters of a linear classifier for the source data.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 107, + 258, + 503, + 292 + ], + "lines": [ + { + "bbox": [ + 106, + 258, + 500, + 271 + ], + "spans": [ + { + "bbox": [ + 106, + 258, + 490, + 271 + ], + "score": 1.0, + "content": "Optimization. To optimize equation 6, we follow a similar alternating strategy as before. The", + "type": "text" + }, + { + "bbox": [ + 490, + 259, + 500, + 268 + ], + "score": 0.75, + "content": "\\textbf { { B } }", + "type": "inline_equation" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 269, + 506, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 269, + 179, + 282 + ], + "score": 1.0, + "content": "minimization and", + "type": "text" + }, + { + "bbox": [ + 179, + 270, + 188, + 280 + ], + "score": 0.79, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 269, + 361, + 282 + ], + "score": 1.0, + "content": "-minimization steps are unchanged, but the", + "type": "text" + }, + { + "bbox": [ + 361, + 270, + 369, + 280 + ], + "score": 0.8, + "content": "\\pmb { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 269, + 506, + 282 + ], + "score": 1.0, + "content": "-minimization now incorporates a", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 281, + 466, + 292 + ], + "spans": [ + { + "bbox": [ + 106, + 281, + 375, + 292 + ], + "score": 1.0, + "content": "new term and we further need to solve for the classifier parameters", + "type": "text" + }, + { + "bbox": [ + 375, + 281, + 388, + 290 + ], + "score": 0.59, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 281, + 466, + 292 + ], + "score": 1.0, + "content": ". This translates to:", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 106, + 297, + 505, + 355 + ], + "lines": [ + { + "bbox": [ + 107, + 296, + 505, + 310 + ], + "spans": [ + { + "bbox": [ + 107, + 298, + 115, + 308 + ], + "score": 0.74, + "content": "\\pmb { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 115, + 296, + 333, + 310 + ], + "score": 1.0, + "content": "-minimization: Minimizing equation 6 with respect to", + "type": "text" + }, + { + "bbox": [ + 333, + 298, + 341, + 307 + ], + "score": 0.79, + "content": "_ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 296, + 505, + 310 + ], + "score": 1.0, + "content": ", with all the other parameters fixed, still", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 308, + 505, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 308, + 378, + 321 + ], + "score": 1.0, + "content": "reduces to a linear least-squares problem. The two terms involving", + "type": "text" + }, + { + "bbox": [ + 379, + 309, + 388, + 319 + ], + "score": 0.78, + "content": "_ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 308, + 505, + 321 + ], + "score": 1.0, + "content": "can be grouped into a single", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 316, + 506, + 346 + ], + "spans": [ + { + "bbox": [ + 104, + 316, + 171, + 345 + ], + "score": 1.0, + "content": "one of the form", + "type": "text" + }, + { + "bbox": [ + 171, + 325, + 255, + 339 + ], + "score": 0.92, + "content": "\\| X _ { n e w } - V _ { n e w } S \\| _ { F } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 316, + 286, + 345 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 286, + 319, + 368, + 345 + ], + "score": 0.95, + "content": "X _ { n e w } = \\binom { \\sqrt { \\alpha } X _ { s } } { \\sqrt { \\beta } L }", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 323, + 387, + 342 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 387, + 319, + 465, + 346 + ], + "score": 0.94, + "content": "V _ { n e w } = \\overset { \\overline { { { \\rho } } } } { \\left( \\overset { \\overline { { { \\alpha } } } } { \\sqrt { \\beta } } W \\right) }", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 318, + 506, + 345 + ], + "score": 1.0, + "content": ", and thus", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 107, + 342, + 243, + 356 + ], + "spans": [ + { + "bbox": [ + 107, + 344, + 115, + 353 + ], + "score": 0.77, + "content": "_ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 116, + 342, + 243, + 356 + ], + "score": 1.0, + "content": "can be obtained in closed form.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14.5 + }, + { + "type": "text", + "bbox": [ + 106, + 360, + 504, + 383 + ], + "lines": [ + { + "bbox": [ + 107, + 358, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 107, + 361, + 120, + 371 + ], + "score": 0.53, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 120, + 358, + 367, + 374 + ], + "score": 1.0, + "content": "-minimization: With all the other parameters fixed, finding", + "type": "text" + }, + { + "bbox": [ + 367, + 361, + 380, + 371 + ], + "score": 0.67, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 358, + 505, + 374 + ], + "score": 1.0, + "content": "corresponds to a linear least-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 371, + 292, + 383 + ], + "spans": [ + { + "bbox": [ + 106, + 371, + 292, + 383 + ], + "score": 1.0, + "content": "squares problem, with a closed-form solution.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 106, + 389, + 505, + 424 + ], + "lines": [ + { + "bbox": [ + 105, + 388, + 504, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 388, + 504, + 402 + ], + "score": 1.0, + "content": "Inference. The same inference strategy as before can be followed to label the target samples.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 400, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 106, + 400, + 262, + 414 + ], + "score": 1.0, + "content": "Another option here is to make use of", + "type": "text" + }, + { + "bbox": [ + 262, + 401, + 276, + 411 + ], + "score": 0.64, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 400, + 505, + 414 + ], + "score": 1.0, + "content": "to classify the samples identified as belonging to known", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 411, + 352, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 411, + 352, + 425 + ], + "score": 1.0, + "content": "classes. We compare these two strategies in our experiments.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20 + }, + { + "type": "title", + "bbox": [ + 107, + 436, + 397, + 448 + ], + "lines": [ + { + "bbox": [ + 106, + 436, + 397, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 436, + 397, + 450 + ], + "score": 1.0, + "content": "3.3 D-FRODA-U: D-FRODA WITH UNKNOWN SOURCE CLASSES", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 457, + 506, + 535 + ], + "lines": [ + { + "bbox": [ + 105, + 457, + 505, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 505, + 470 + ], + "score": 1.0, + "content": "Until now, we have tackled the scenario where there are no unknown classes in the source data,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 469, + 506, + 482 + ], + "spans": [ + { + "bbox": [ + 106, + 469, + 506, + 482 + ], + "score": 1.0, + "content": "which we believe corresponds to the typical application scenario, since the source data can in general", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 479, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 505, + 493 + ], + "score": 1.0, + "content": "be fully annotated. Nevertheless, to match the scenario of Busto & Gall (2017), who assume to have", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 104, + 490, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 104, + 490, + 505, + 503 + ], + "score": 1.0, + "content": "access to additional source samples from unknown classes, yet different from the target unknown", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 501, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 505, + 514 + ], + "score": 1.0, + "content": "classes, we introduce a modified version of our approach that takes such auxiliary data into account.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 513, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 505, + 525 + ], + "score": 1.0, + "content": "Note that, since one knows which source samples are from unknown classes, it is also possible to", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 524, + 468, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 468, + 536 + ], + "score": 1.0, + "content": "simply discard them from training. To nonetheless handle them, we re-write equation 6 as", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 26 + }, + { + "type": "interline_equation", + "bbox": [ + 137, + 538, + 442, + 631 + ], + "lines": [ + { + "bbox": [ + 137, + 538, + 442, + 631 + ], + "spans": [ + { + "bbox": [ + 137, + 538, + 442, + 631 + ], + "score": 0.94, + "content": "\\begin{array} { r l } { \\underset { U , T , V , U ^ { \\prime } , S ^ { \\prime } , W ^ { \\prime } } { \\operatorname* { m i n } } } & { \\| X _ { t } - B \\pmb { T } \\| _ { F } ^ { 2 } + \\alpha \\| X ^ { \\prime } _ { s } - B ^ { \\prime } S ^ { \\prime } \\| _ { F } ^ { 2 } + \\beta \\| \\pmb { L } - W ^ { \\prime } S ^ { \\prime } \\| _ { F } ^ { 2 } } \\\\ & { + \\lambda _ { 1 } \\displaystyle \\sum _ { i = 1 } ^ { n _ { t } } ( \\| \\pmb { T } _ { i } ^ { v } \\| + \\| \\pmb { T } _ { i } ^ { u } \\| ) + \\lambda _ { 2 } \\displaystyle \\sum _ { i = 1 } ^ { n _ { s } } \\big ( \\| \\pmb { S } ^ { \\prime } _ { i } ^ { v } \\| + \\| \\pmb { S } ^ { \\prime } _ { i } ^ { u } \\| \\big ) } \\\\ & { \\qquad \\quad \\ : s . t . \\quad \\displaystyle \\sum _ { j = 1 } ^ { 2 d } \\| \\pmb { B } _ { j } \\| ^ { 2 } \\leq 1 , \\displaystyle \\sum _ { j = 1 } ^ { 2 d } \\| \\pmb { B } ^ { \\prime } _ { j } \\| ^ { 2 } \\leq 1 , } \\end{array}", + "type": "interline_equation", + "image_path": "a26dca4589c4287166db7102109a80b7876b04b5045729b931a2134125559640.jpg" + } + ] + } + ], + "index": 31, + "virtual_lines": [ + { + "bbox": [ + 137, + 538, + 442, + 569.0 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 137, + 569.0, + 442, + 600.0 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 137, + 600.0, + 442, + 631.0 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 635, + 505, + 703 + ], + "lines": [ + { + "bbox": [ + 106, + 635, + 504, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 635, + 136, + 648 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 136, + 636, + 154, + 647 + ], + "score": 0.88, + "content": "X ^ { \\prime } { } _ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 635, + 475, + 648 + ], + "score": 1.0, + "content": "contains the source samples from both known and unknown classes, and", + "type": "text" + }, + { + "bbox": [ + 475, + 636, + 504, + 647 + ], + "score": 0.87, + "content": "\\mathbf { { } \\delta } B ^ { \\prime } \\mathbf { \\delta } =", + "type": "inline_equation" + } + ], + "index": 33 + }, + { + "bbox": [ + 107, + 644, + 507, + 661 + ], + "spans": [ + { + "bbox": [ + 107, + 647, + 185, + 659 + ], + "score": 0.9, + "content": "[ V , U ^ { \\prime } ] \\in \\mathbb { R } ^ { D \\times 2 d }", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 644, + 381, + 661 + ], + "score": 1.0, + "content": "denotes the source transformation matrix with", + "type": "text" + }, + { + "bbox": [ + 381, + 646, + 436, + 658 + ], + "score": 0.92, + "content": "U ^ { \\prime } \\in \\mathbb { R } ^ { D \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 644, + 507, + 661 + ], + "score": 1.0, + "content": "the private sub-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 657, + 506, + 671 + ], + "spans": [ + { + "bbox": [ + 104, + 657, + 506, + 671 + ], + "score": 1.0, + "content": "space for the source data. Note that, similarly to the target coefficients, we have now separated the", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 666, + 508, + 684 + ], + "spans": [ + { + "bbox": [ + 104, + 666, + 250, + 684 + ], + "score": 1.0, + "content": "source coefficients for each sample", + "type": "text" + }, + { + "bbox": [ + 250, + 669, + 265, + 680 + ], + "score": 0.88, + "content": "{ \\mathbf { } } S _ { \\mathrm { ~ } { i } } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 265, + 666, + 463, + 684 + ], + "score": 1.0, + "content": "into a part corresponding to the shared subspace", + "type": "text" + }, + { + "bbox": [ + 463, + 669, + 479, + 681 + ], + "score": 0.94, + "content": "{ \\mathbf { } } S _ { \\textit { i } } ^ { \\prime \\ v }", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 666, + 508, + 684 + ], + "score": 1.0, + "content": "and a", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 103, + 677, + 507, + 695 + ], + "spans": [ + { + "bbox": [ + 103, + 677, + 256, + 695 + ], + "score": 1.0, + "content": "part corresponding to the private one", + "type": "text" + }, + { + "bbox": [ + 256, + 681, + 272, + 692 + ], + "score": 0.9, + "content": "{ S ^ { \\prime } } _ { i } ^ { u }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 677, + 434, + 695 + ], + "score": 1.0, + "content": ". Note also that the classifier parameters", + "type": "text" + }, + { + "bbox": [ + 434, + 680, + 451, + 690 + ], + "score": 0.83, + "content": "W ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 677, + 507, + 695 + ], + "score": 1.0, + "content": "now account", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 690, + 420, + 704 + ], + "spans": [ + { + "bbox": [ + 105, + 690, + 120, + 704 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 121, + 691, + 147, + 702 + ], + "score": 0.9, + "content": "C + 1", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 690, + 420, + 704 + ], + "score": 1.0, + "content": "classes, the additional class corresponding to the unknown samples.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Optimization. We follow a similar iterative procedure to the one used before, with modifications", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 720, + 273, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 146, + 732 + ], + "score": 1.0, + "content": "to update", + "type": "text" + }, + { + "bbox": [ + 146, + 721, + 159, + 731 + ], + "score": 0.86, + "content": "B ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 720, + 177, + 732 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 178, + 721, + 189, + 731 + ], + "score": 0.85, + "content": "S ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 720, + 273, + 732 + ], + "score": 1.0, + "content": ", as discussed below.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 82, + 294, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 295, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 295, + 95 + ], + "score": 1.0, + "content": "3.2 D-FRODA: DISCRIMINATIVE FRODA", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 102, + 505, + 159 + ], + "lines": [ + { + "bbox": [ + 106, + 102, + 505, + 116 + ], + "spans": [ + { + "bbox": [ + 106, + 102, + 505, + 116 + ], + "score": 1.0, + "content": "The formulation above does not make use of the source labels at all during the representation learn-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 114, + 505, + 126 + ], + "spans": [ + { + "bbox": [ + 106, + 114, + 505, + 126 + ], + "score": 1.0, + "content": "ing stage. As such, it does not encourage the representation to be discriminative. To overcome this,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 123, + 505, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 123, + 505, + 139 + ], + "score": 1.0, + "content": "we extend our basic formulation to further account for the classification task at hand. Specifically,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 134, + 506, + 150 + ], + "spans": [ + { + "bbox": [ + 104, + 134, + 119, + 150 + ], + "score": 1.0, + "content": "let", + "type": "text" + }, + { + "bbox": [ + 119, + 135, + 224, + 148 + ], + "score": 0.92, + "content": "\\pmb { L } = [ l _ { 1 } \\dots l _ { n _ { s } } ] \\in \\mathbb { R } ^ { C \\times n _ { s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 134, + 425, + 150 + ], + "score": 1.0, + "content": "be the matrix containing the source labels, where", + "type": "text" + }, + { + "bbox": [ + 426, + 136, + 460, + 147 + ], + "score": 0.92, + "content": "\\boldsymbol { l } _ { i } \\in \\mathbb { R } ^ { C }", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 134, + 506, + 150 + ], + "score": 1.0, + "content": "represents", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 147, + 471, + 159 + ], + "spans": [ + { + "bbox": [ + 106, + 147, + 282, + 159 + ], + "score": 1.0, + "content": "the one-hot encoding of the label of sample", + "type": "text" + }, + { + "bbox": [ + 282, + 148, + 286, + 157 + ], + "score": 0.68, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 147, + 471, + 159 + ], + "score": 1.0, + "content": ". We then write our D-FRODA formulation as", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3, + "bbox_fs": [ + 104, + 102, + 506, + 159 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 112, + 164, + 488, + 234 + ], + "lines": [ + { + "bbox": [ + 112, + 164, + 488, + 234 + ], + "spans": [ + { + "bbox": [ + 112, + 164, + 488, + 234 + ], + "score": 0.93, + "content": "\\begin{array} { r l } { \\displaystyle \\underset { U , T , V , S , W } { \\operatorname* { m i n } } } & { \\| X _ { t } - B T \\| _ { F } ^ { 2 } + \\alpha \\| X _ { s } - V S \\| _ { F } ^ { 2 } + \\beta \\| L - W S \\| _ { F } ^ { 2 } + \\lambda _ { 1 } \\displaystyle \\sum _ { i = 1 } ^ { n _ { t } } ( \\| T _ { i } ^ { v } \\| + \\| T _ { i } ^ { u } \\| ) ) } \\\\ { \\displaystyle s . t . } & { \\displaystyle \\displaystyle \\sum _ { j = 1 } ^ { d } \\| U _ { j } \\| ^ { 2 } \\leq 1 , \\displaystyle \\sum _ { j = 1 } ^ { d } \\| V _ { j } \\| ^ { 2 } \\leq 1 , } \\end{array}", + "type": "interline_equation", + "image_path": "a4cec1062f3cabe4da7a316f96ab8b73126525037273f676a43ae2a72aa32461.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 112, + 164, + 488, + 187.33333333333334 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 112, + 187.33333333333334, + 488, + 210.66666666666669 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 112, + 210.66666666666669, + 488, + 234.00000000000003 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 239, + 502, + 252 + ], + "lines": [ + { + "bbox": [ + 105, + 236, + 503, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 236, + 133, + 254 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 239, + 183, + 250 + ], + "score": 0.92, + "content": "W \\in \\mathbb { R } ^ { C \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 236, + 503, + 254 + ], + "score": 1.0, + "content": "is the matrix containing the parameters of a linear classifier for the source data.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 236, + 503, + 254 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 258, + 503, + 292 + ], + "lines": [ + { + "bbox": [ + 106, + 258, + 500, + 271 + ], + "spans": [ + { + "bbox": [ + 106, + 258, + 490, + 271 + ], + "score": 1.0, + "content": "Optimization. To optimize equation 6, we follow a similar alternating strategy as before. The", + "type": "text" + }, + { + "bbox": [ + 490, + 259, + 500, + 268 + ], + "score": 0.75, + "content": "\\textbf { { B } }", + "type": "inline_equation" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 269, + 506, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 269, + 179, + 282 + ], + "score": 1.0, + "content": "minimization and", + "type": "text" + }, + { + "bbox": [ + 179, + 270, + 188, + 280 + ], + "score": 0.79, + "content": "_ { \\mathbf { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 269, + 361, + 282 + ], + "score": 1.0, + "content": "-minimization steps are unchanged, but the", + "type": "text" + }, + { + "bbox": [ + 361, + 270, + 369, + 280 + ], + "score": 0.8, + "content": "\\pmb { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 269, + 506, + 282 + ], + "score": 1.0, + "content": "-minimization now incorporates a", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 281, + 466, + 292 + ], + "spans": [ + { + "bbox": [ + 106, + 281, + 375, + 292 + ], + "score": 1.0, + "content": "new term and we further need to solve for the classifier parameters", + "type": "text" + }, + { + "bbox": [ + 375, + 281, + 388, + 290 + ], + "score": 0.59, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 281, + 466, + 292 + ], + "score": 1.0, + "content": ". This translates to:", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11, + "bbox_fs": [ + 105, + 258, + 506, + 292 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 297, + 505, + 355 + ], + "lines": [ + { + "bbox": [ + 107, + 296, + 505, + 310 + ], + "spans": [ + { + "bbox": [ + 107, + 298, + 115, + 308 + ], + "score": 0.74, + "content": "\\pmb { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 115, + 296, + 333, + 310 + ], + "score": 1.0, + "content": "-minimization: Minimizing equation 6 with respect to", + "type": "text" + }, + { + "bbox": [ + 333, + 298, + 341, + 307 + ], + "score": 0.79, + "content": "_ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 296, + 505, + 310 + ], + "score": 1.0, + "content": ", with all the other parameters fixed, still", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 308, + 505, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 308, + 378, + 321 + ], + "score": 1.0, + "content": "reduces to a linear least-squares problem. The two terms involving", + "type": "text" + }, + { + "bbox": [ + 379, + 309, + 388, + 319 + ], + "score": 0.78, + "content": "_ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 308, + 505, + 321 + ], + "score": 1.0, + "content": "can be grouped into a single", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 316, + 506, + 346 + ], + "spans": [ + { + "bbox": [ + 104, + 316, + 171, + 345 + ], + "score": 1.0, + "content": "one of the form", + "type": "text" + }, + { + "bbox": [ + 171, + 325, + 255, + 339 + ], + "score": 0.92, + "content": "\\| X _ { n e w } - V _ { n e w } S \\| _ { F } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 316, + 286, + 345 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 286, + 319, + 368, + 345 + ], + "score": 0.95, + "content": "X _ { n e w } = \\binom { \\sqrt { \\alpha } X _ { s } } { \\sqrt { \\beta } L }", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 323, + 387, + 342 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 387, + 319, + 465, + 346 + ], + "score": 0.94, + "content": "V _ { n e w } = \\overset { \\overline { { { \\rho } } } } { \\left( \\overset { \\overline { { { \\alpha } } } } { \\sqrt { \\beta } } W \\right) }", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 318, + 506, + 345 + ], + "score": 1.0, + "content": ", and thus", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 107, + 342, + 243, + 356 + ], + "spans": [ + { + "bbox": [ + 107, + 344, + 115, + 353 + ], + "score": 0.77, + "content": "_ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 116, + 342, + 243, + 356 + ], + "score": 1.0, + "content": "can be obtained in closed form.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14.5, + "bbox_fs": [ + 104, + 296, + 506, + 356 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 360, + 504, + 383 + ], + "lines": [ + { + "bbox": [ + 107, + 358, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 107, + 361, + 120, + 371 + ], + "score": 0.53, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 120, + 358, + 367, + 374 + ], + "score": 1.0, + "content": "-minimization: With all the other parameters fixed, finding", + "type": "text" + }, + { + "bbox": [ + 367, + 361, + 380, + 371 + ], + "score": 0.67, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 358, + 505, + 374 + ], + "score": 1.0, + "content": "corresponds to a linear least-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 371, + 292, + 383 + ], + "spans": [ + { + "bbox": [ + 106, + 371, + 292, + 383 + ], + "score": 1.0, + "content": "squares problem, with a closed-form solution.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 106, + 358, + 505, + 383 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 389, + 505, + 424 + ], + "lines": [ + { + "bbox": [ + 105, + 388, + 504, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 388, + 504, + 402 + ], + "score": 1.0, + "content": "Inference. The same inference strategy as before can be followed to label the target samples.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 400, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 106, + 400, + 262, + 414 + ], + "score": 1.0, + "content": "Another option here is to make use of", + "type": "text" + }, + { + "bbox": [ + 262, + 401, + 276, + 411 + ], + "score": 0.64, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 400, + 505, + 414 + ], + "score": 1.0, + "content": "to classify the samples identified as belonging to known", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 411, + 352, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 411, + 352, + 425 + ], + "score": 1.0, + "content": "classes. We compare these two strategies in our experiments.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 388, + 505, + 425 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 436, + 397, + 448 + ], + "lines": [ + { + "bbox": [ + 106, + 436, + 397, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 436, + 397, + 450 + ], + "score": 1.0, + "content": "3.3 D-FRODA-U: D-FRODA WITH UNKNOWN SOURCE CLASSES", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 457, + 506, + 535 + ], + "lines": [ + { + "bbox": [ + 105, + 457, + 505, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 505, + 470 + ], + "score": 1.0, + "content": "Until now, we have tackled the scenario where there are no unknown classes in the source data,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 469, + 506, + 482 + ], + "spans": [ + { + "bbox": [ + 106, + 469, + 506, + 482 + ], + "score": 1.0, + "content": "which we believe corresponds to the typical application scenario, since the source data can in general", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 479, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 505, + 493 + ], + "score": 1.0, + "content": "be fully annotated. Nevertheless, to match the scenario of Busto & Gall (2017), who assume to have", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 104, + 490, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 104, + 490, + 505, + 503 + ], + "score": 1.0, + "content": "access to additional source samples from unknown classes, yet different from the target unknown", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 501, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 505, + 514 + ], + "score": 1.0, + "content": "classes, we introduce a modified version of our approach that takes such auxiliary data into account.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 513, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 505, + 525 + ], + "score": 1.0, + "content": "Note that, since one knows which source samples are from unknown classes, it is also possible to", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 524, + 468, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 468, + 536 + ], + "score": 1.0, + "content": "simply discard them from training. To nonetheless handle them, we re-write equation 6 as", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 26, + "bbox_fs": [ + 104, + 457, + 506, + 536 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 137, + 538, + 442, + 631 + ], + "lines": [ + { + "bbox": [ + 137, + 538, + 442, + 631 + ], + "spans": [ + { + "bbox": [ + 137, + 538, + 442, + 631 + ], + "score": 0.94, + "content": "\\begin{array} { r l } { \\underset { U , T , V , U ^ { \\prime } , S ^ { \\prime } , W ^ { \\prime } } { \\operatorname* { m i n } } } & { \\| X _ { t } - B \\pmb { T } \\| _ { F } ^ { 2 } + \\alpha \\| X ^ { \\prime } _ { s } - B ^ { \\prime } S ^ { \\prime } \\| _ { F } ^ { 2 } + \\beta \\| \\pmb { L } - W ^ { \\prime } S ^ { \\prime } \\| _ { F } ^ { 2 } } \\\\ & { + \\lambda _ { 1 } \\displaystyle \\sum _ { i = 1 } ^ { n _ { t } } ( \\| \\pmb { T } _ { i } ^ { v } \\| + \\| \\pmb { T } _ { i } ^ { u } \\| ) + \\lambda _ { 2 } \\displaystyle \\sum _ { i = 1 } ^ { n _ { s } } \\big ( \\| \\pmb { S } ^ { \\prime } _ { i } ^ { v } \\| + \\| \\pmb { S } ^ { \\prime } _ { i } ^ { u } \\| \\big ) } \\\\ & { \\qquad \\quad \\ : s . t . \\quad \\displaystyle \\sum _ { j = 1 } ^ { 2 d } \\| \\pmb { B } _ { j } \\| ^ { 2 } \\leq 1 , \\displaystyle \\sum _ { j = 1 } ^ { 2 d } \\| \\pmb { B } ^ { \\prime } _ { j } \\| ^ { 2 } \\leq 1 , } \\end{array}", + "type": "interline_equation", + "image_path": "a26dca4589c4287166db7102109a80b7876b04b5045729b931a2134125559640.jpg" + } + ] + } + ], + "index": 31, + "virtual_lines": [ + { + "bbox": [ + 137, + 538, + 442, + 569.0 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 137, + 569.0, + 442, + 600.0 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 137, + 600.0, + 442, + 631.0 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 635, + 505, + 703 + ], + "lines": [ + { + "bbox": [ + 106, + 635, + 504, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 635, + 136, + 648 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 136, + 636, + 154, + 647 + ], + "score": 0.88, + "content": "X ^ { \\prime } { } _ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 635, + 475, + 648 + ], + "score": 1.0, + "content": "contains the source samples from both known and unknown classes, and", + "type": "text" + }, + { + "bbox": [ + 475, + 636, + 504, + 647 + ], + "score": 0.87, + "content": "\\mathbf { { } \\delta } B ^ { \\prime } \\mathbf { \\delta } =", + "type": "inline_equation" + } + ], + "index": 33 + }, + { + "bbox": [ + 107, + 644, + 507, + 661 + ], + "spans": [ + { + "bbox": [ + 107, + 647, + 185, + 659 + ], + "score": 0.9, + "content": "[ V , U ^ { \\prime } ] \\in \\mathbb { R } ^ { D \\times 2 d }", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 644, + 381, + 661 + ], + "score": 1.0, + "content": "denotes the source transformation matrix with", + "type": "text" + }, + { + "bbox": [ + 381, + 646, + 436, + 658 + ], + "score": 0.92, + "content": "U ^ { \\prime } \\in \\mathbb { R } ^ { D \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 644, + 507, + 661 + ], + "score": 1.0, + "content": "the private sub-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 657, + 506, + 671 + ], + "spans": [ + { + "bbox": [ + 104, + 657, + 506, + 671 + ], + "score": 1.0, + "content": "space for the source data. Note that, similarly to the target coefficients, we have now separated the", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 666, + 508, + 684 + ], + "spans": [ + { + "bbox": [ + 104, + 666, + 250, + 684 + ], + "score": 1.0, + "content": "source coefficients for each sample", + "type": "text" + }, + { + "bbox": [ + 250, + 669, + 265, + 680 + ], + "score": 0.88, + "content": "{ \\mathbf { } } S _ { \\mathrm { ~ } { i } } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 265, + 666, + 463, + 684 + ], + "score": 1.0, + "content": "into a part corresponding to the shared subspace", + "type": "text" + }, + { + "bbox": [ + 463, + 669, + 479, + 681 + ], + "score": 0.94, + "content": "{ \\mathbf { } } S _ { \\textit { i } } ^ { \\prime \\ v }", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 666, + 508, + 684 + ], + "score": 1.0, + "content": "and a", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 103, + 677, + 507, + 695 + ], + "spans": [ + { + "bbox": [ + 103, + 677, + 256, + 695 + ], + "score": 1.0, + "content": "part corresponding to the private one", + "type": "text" + }, + { + "bbox": [ + 256, + 681, + 272, + 692 + ], + "score": 0.9, + "content": "{ S ^ { \\prime } } _ { i } ^ { u }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 677, + 434, + 695 + ], + "score": 1.0, + "content": ". Note also that the classifier parameters", + "type": "text" + }, + { + "bbox": [ + 434, + 680, + 451, + 690 + ], + "score": 0.83, + "content": "W ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 677, + 507, + 695 + ], + "score": 1.0, + "content": "now account", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 690, + 420, + 704 + ], + "spans": [ + { + "bbox": [ + 105, + 690, + 120, + 704 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 121, + 691, + 147, + 702 + ], + "score": 0.9, + "content": "C + 1", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 690, + 420, + 704 + ], + "score": 1.0, + "content": "classes, the additional class corresponding to the unknown samples.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 35.5, + "bbox_fs": [ + 103, + 635, + 508, + 704 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Optimization. We follow a similar iterative procedure to the one used before, with modifications", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 720, + 273, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 146, + 732 + ], + "score": 1.0, + "content": "to update", + "type": "text" + }, + { + "bbox": [ + 146, + 721, + 159, + 731 + ], + "score": 0.86, + "content": "B ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 720, + 177, + 732 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 178, + 721, + 189, + 731 + ], + "score": 0.85, + "content": "S ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 720, + 273, + 732 + ], + "score": 1.0, + "content": ", as discussed below.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5, + "bbox_fs": [ + 105, + 709, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 108, + 105, + 510, + 276 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 80, + 504, + 101 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 79, + 506, + 92 + ], + "spans": [ + { + "bbox": [ + 105, + 79, + 506, + 92 + ], + "score": 1.0, + "content": "Table 1: Recognition accuracies on the 12 source/target pairs of the BCIS dataset (Tommasi & Tuytelaars,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 90, + 407, + 101 + ], + "spans": [ + { + "bbox": [ + 106, + 90, + 407, + 101 + ], + "score": 1.0, + "content": "2014) using a linear SVM classifier. B: Bing, C: Caltech256, I: ImageNet, S: SUN.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table_body", + "bbox": [ + 108, + 105, + 510, + 276 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 105, + 510, + 276 + ], + "spans": [ + { + "bbox": [ + 108, + 105, + 510, + 276 + ], + "score": 0.98, + "html": "
MethodB→CB→IB→SC→BC→IC→S
TCA (Pan et al.,2011) GFK (Gong et al., 2012)62.8±3.8 56.6± 4.5 66.2 ± 4.0 58.3 ±3.129.6± 4.2 23.8 ±2.038.9±1.9 40.2 ± 1.860.2 ± 1.4 62.2 ± 1.529.7± 1.6 28.5 ± 1.0
SA (Fernando et al.,2013) CORAL (Sun et al., 2016)66.0±3.4 57.8±3.224.3 ± 2.640.3 ± 1.762.5 ± 0.829.0 ± 1.5
ATI (Busto & Gall, 2017)68.8 ±3.3 60.9 ± 2.627.2 ±3.940.7 ± 1.564.0 ± 2.631.4±0.8
AODA (Saito et al., 2018)71.4 ± 2.3 69.0±2.8 76.2 ±1.7 70.9 ±3.237.4± 2.6 57.3 ± 1.145.7 ± 3.0 63.5 ± 2.167.9 ± 4.2 73.5 ± 0.837.5 ± 2.7 60.5±0.8
FRODA D-FRODA73.8±6.1 71.0± 2.0 74.6 ± 5.5 71.4± 2.054.7 ± 2.9 55.4± 2.767.5 ± 1.4 67.6 ± 1.274.5 ± 1.7 75.0±1.861.6±2.2 61.7 ± 2.1
Method TCA (Pan et al.,2011)I→B I→C 40.9±2.9 68.6±1.8I→S 34.5±3.8S→B 19.4 ± 2.1S→C 32.0±3.9S→I Avg. 31.1 ± 4.6 42±3.04
GFK(Gong et al.,2012) SA (Fernando et al.,2013) CORAL (Sun et al.,2016)42.6 ± 2.4 73.3± 3.6 43.1 ± 1.6 72.8 ± 3.1 44.6 ± 2.532.7± 3.6 32.2±3.716.9 ± 1.5 17.5 ± 1.628.6±3.8 26.4± 1.1 29.2 ± 4.2 27.1 ± 1.341.6 ± 2.5 41.8 ± 2.4
ATI (Busto & Gall,2017) AODA (Saito et al., 2018)74.5 ± 3.4 48.8±2.3 77.5 ± 2.2 66.3± 0.9 78.1± 0.935.4 ± 4.4 43.4± 4.818.7 ± 1.2 23.2 ±3.233.6 ± 5.3 31.3 ± 1.3 47.3±2.944.3 ± 2.7 50.2 ± 2.8
FRODA D-FRODA66.0±1.9 79.9 ± 1.7 66.4±1.7 80.5 ± 1.659.4± 1.4 59.2± 2.1 55.7 ± 2.5 59.8 ±2.0 55.5 ± 2.456.5 ± 2.633.0 ±1.1 59.6 ±3.1 63.2 ± 1.3 61.2 ± 1.8 59.4± 1.965.4 ±1.7 65.4± 2.3
", + "type": "table", + "image_path": "8ee7e9346a61f58f347d8877435cc5a9ffe24cc2ce8caac2212b2230efb43686.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 108, + 105, + 510, + 162.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 108, + 162.0, + 510, + 219.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 108, + 219.0, + 510, + 276.0 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 1.75 + }, + { + "type": "text", + "bbox": [ + 106, + 286, + 479, + 298 + ], + "lines": [ + { + "bbox": [ + 107, + 285, + 479, + 299 + ], + "spans": [ + { + "bbox": [ + 107, + 286, + 118, + 296 + ], + "score": 0.84, + "content": "S ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 118, + 285, + 299, + 299 + ], + "score": 1.0, + "content": "-minimization: To minimize equation 7 w.r.t.", + "type": "text" + }, + { + "bbox": [ + 299, + 286, + 311, + 297 + ], + "score": 0.86, + "content": "S ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 285, + 479, + 299 + ], + "score": 1.0, + "content": ", with all other parameters fixed, we write", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "interline_equation", + "bbox": [ + 151, + 300, + 457, + 333 + ], + "lines": [ + { + "bbox": [ + 151, + 300, + 457, + 333 + ], + "spans": [ + { + "bbox": [ + 151, + 300, + 457, + 333 + ], + "score": 0.93, + "content": "\\operatorname* { m i n } _ { { \\boldsymbol { S } } ^ { \\prime } } \\quad \\alpha \\| { \\boldsymbol { X } } ^ { \\prime } { \\boldsymbol { s } } - { \\boldsymbol { B } } ^ { \\prime } { \\boldsymbol { S } } ^ { \\prime } \\| _ { F } ^ { 2 } + \\beta \\| { \\boldsymbol { L } } - { \\boldsymbol { W } } ^ { \\prime } { \\boldsymbol { S } } ^ { \\prime } \\| _ { F } ^ { 2 } + \\lambda _ { 2 } \\sum _ { i = 1 } ^ { n _ { s } } \\left( \\| { \\boldsymbol { S } } ^ { \\prime } { \\boldsymbol { s } } ^ { \\prime } \\| + \\| { \\boldsymbol { S } } ^ { \\prime } { \\boldsymbol { u } } \\| \\right) ~ .", + "type": "interline_equation", + "image_path": "83aa450125b6ed8ad171484ca199a1b14a521bc1902386d2cdec09b3ef04749a.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 151, + 300, + 457, + 311.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 151, + 311.0, + 457, + 322.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 151, + 322.0, + 457, + 333.0 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 334, + 505, + 368 + ], + "lines": [ + { + "bbox": [ + 105, + 333, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 505, + 348 + ], + "score": 1.0, + "content": "The first two terms can be grouped into a single squared Frobenius norm, thus resulting in a sparse", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 345, + 505, + 359 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 298, + 359 + ], + "score": 1.0, + "content": "group lasso problem, which, as when updating", + "type": "text" + }, + { + "bbox": [ + 298, + 346, + 307, + 356 + ], + "score": 0.81, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 345, + 505, + 359 + ], + "score": 1.0, + "content": "in FRODA, can be solved via proximal gradient", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 356, + 223, + 369 + ], + "spans": [ + { + "bbox": [ + 106, + 356, + 223, + 369 + ], + "score": 1.0, + "content": "descent (Mairal et al., 2014).", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 108, + 372, + 444, + 385 + ], + "lines": [ + { + "bbox": [ + 107, + 369, + 446, + 389 + ], + "spans": [ + { + "bbox": [ + 107, + 373, + 120, + 383 + ], + "score": 0.84, + "content": "B ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 120, + 369, + 271, + 389 + ], + "score": 1.0, + "content": "-minimization: Given the coefficients", + "type": "text" + }, + { + "bbox": [ + 271, + 373, + 287, + 383 + ], + "score": 0.89, + "content": "S ^ { \\prime ^ { v } }", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 369, + 305, + 389 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 306, + 373, + 322, + 383 + ], + "score": 0.89, + "content": "S ^ { \\prime } { } ^ { u }", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 369, + 385, + 389 + ], + "score": 1.0, + "content": ", we can update", + "type": "text" + }, + { + "bbox": [ + 385, + 374, + 399, + 383 + ], + "score": 0.87, + "content": "U ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 369, + 446, + 389 + ], + "score": 1.0, + "content": "by solving", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "interline_equation", + "bbox": [ + 210, + 387, + 399, + 423 + ], + "lines": [ + { + "bbox": [ + 210, + 387, + 399, + 423 + ], + "spans": [ + { + "bbox": [ + 210, + 387, + 399, + 423 + ], + "score": 0.94, + "content": "\\operatorname* { m i n } _ { U ^ { \\prime } } \\quad \\| A - U ^ { \\prime } S ^ { \\prime ^ { \\# } } \\| _ { F } ^ { 2 } \\quad s . t . \\sum _ { j = 1 } ^ { d } \\| U ^ { \\prime } { } _ { j } \\| ^ { 2 } \\leq 1 ,", + "type": "interline_equation", + "image_path": "e064d3974e5e9c0fb0c605063d71cb08f06bd13e992196deb75d0d3a0109b414.jpg" + } + ] + } + ], + "index": 13.5, + "virtual_lines": [ + { + "bbox": [ + 210, + 387, + 399, + 405.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 210, + 405.0, + 399, + 423.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 425, + 281, + 438 + ], + "lines": [ + { + "bbox": [ + 105, + 422, + 283, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 422, + 127, + 442 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 127, + 425, + 204, + 438 + ], + "score": 0.92, + "content": "A = { X ^ { \\prime } } _ { s } - V { S ^ { \\prime } } ^ { v }", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 422, + 225, + 442 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 225, + 426, + 235, + 436 + ], + "score": 0.82, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 422, + 283, + 442 + ], + "score": 1.0, + "content": "by solving", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "interline_equation", + "bbox": [ + 216, + 440, + 393, + 475 + ], + "lines": [ + { + "bbox": [ + 216, + 440, + 393, + 475 + ], + "spans": [ + { + "bbox": [ + 216, + 440, + 393, + 475 + ], + "score": 0.94, + "content": "\\operatorname* { m i n } _ { V } \\quad \\| A ^ { \\prime } - V C \\| _ { F } ^ { 2 } \\quad s . t . \\sum _ { j = 1 } ^ { d } \\| V _ { j } \\| ^ { 2 } \\leq 1 ,", + "type": "interline_equation", + "image_path": "e11cb5d6fcdf5370381a318c998557aef3776053c73c21d1ac6ad328e9b12fe8.jpg" + } + ] + } + ], + "index": 16.5, + "virtual_lines": [ + { + "bbox": [ + 216, + 440, + 393, + 457.5 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 216, + 457.5, + 393, + 475.0 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 478, + 505, + 515 + ], + "lines": [ + { + "bbox": [ + 104, + 468, + 514, + 511 + ], + "spans": [ + { + "bbox": [ + 104, + 479, + 127, + 500 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 127, + 478, + 250, + 504 + ], + "score": 0.93, + "content": "A ^ { \\prime } = \\binom { X _ { t } - U T ^ { u } } { \\sqrt { \\alpha } ( X ^ { \\prime } { } _ { s } - U ^ { \\prime } S ^ { \\prime } { } ^ { u } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 468, + 269, + 511 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 270, + 478, + 338, + 505 + ], + "score": 0.94, + "content": "C = \\left( \\underset { \\sqrt { \\alpha } S ^ { \\prime } } { \\mathbf { \\nabla } } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 468, + 514, + 511 + ], + "score": 1.0, + "content": ". As in FRODA, these two sub-problems", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 500, + 436, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 436, + 516 + ], + "score": 1.0, + "content": "can be solved efficiently using the Lagrange dual formulation of Lee et al. (2007).", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "title", + "bbox": [ + 107, + 527, + 201, + 540 + ], + "lines": [ + { + "bbox": [ + 104, + 525, + 202, + 542 + ], + "spans": [ + { + "bbox": [ + 104, + 525, + 202, + 542 + ], + "score": 1.0, + "content": "4 EXPERIMENTS", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 106, + 545, + 505, + 601 + ], + "lines": [ + { + "bbox": [ + 106, + 546, + 505, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 505, + 558 + ], + "score": 1.0, + "content": "We evaluate our approach on the task of open-set visual domain adaptation using two benchmark", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 556, + 504, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 556, + 504, + 569 + ], + "score": 1.0, + "content": "datasets, and compare its performance against the state-of-the-art open-set domain adaptation meth-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 568, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 106, + 568, + 505, + 579 + ], + "score": 1.0, + "content": "ods on each dataset.1 Note that we also report the results of the methods used as baselines in (Busto", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 578, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 106, + 578, + 246, + 591 + ], + "score": 1.0, + "content": "& Gall, 2017). For a dataset with", + "type": "text" + }, + { + "bbox": [ + 246, + 579, + 256, + 588 + ], + "score": 0.8, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 578, + 428, + 591 + ], + "score": 1.0, + "content": "source classes, we report the accuracy on", + "type": "text" + }, + { + "bbox": [ + 428, + 579, + 455, + 589 + ], + "score": 0.9, + "content": "C + 1", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 578, + 505, + 591 + ], + "score": 1.0, + "content": "classes, the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 588, + 311, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 311, + 603 + ], + "score": 1.0, + "content": "additional one corresponding to the unknown case.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 106, + 603, + 505, + 714 + ], + "lines": [ + { + "bbox": [ + 105, + 603, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 505, + 617 + ], + "score": 1.0, + "content": "Implementation details. Following Busto & Gall (2017), we represent the source and target sam-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 615, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 220, + 627 + ], + "score": 1.0, + "content": "ples with 4096-dimensional", + "type": "text" + }, + { + "bbox": [ + 221, + 615, + 261, + 626 + ], + "score": 0.92, + "content": "D e C A F _ { 7 }", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 615, + 505, + 627 + ], + "score": 1.0, + "content": "features (Donahue et al., 2013) and first reduce their dimen-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 626, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 505, + 639 + ], + "score": 1.0, + "content": "sionality by performing PCA jointly on the source and target data and keeping the components", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 636, + 506, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 145, + 650 + ], + "score": 1.0, + "content": "encoding", + "type": "text" + }, + { + "bbox": [ + 146, + 637, + 166, + 648 + ], + "score": 0.86, + "content": "9 9 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 636, + 406, + 650 + ], + "score": 1.0, + "content": "of the data variance. To then determine the dimensionality", + "type": "text" + }, + { + "bbox": [ + 407, + 637, + 414, + 647 + ], + "score": 0.74, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 636, + 506, + 650 + ], + "score": 1.0, + "content": "of our shared and pri-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 648, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 505, + 660 + ], + "score": 1.0, + "content": "vate subspaces, we make use of the subspace disagreement measure of (Gong et al., 2012). For all", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 659, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 417, + 672 + ], + "score": 1.0, + "content": "our experiments, the hyperparameters of our approach were set as follows:", + "type": "text" + }, + { + "bbox": [ + 417, + 659, + 454, + 669 + ], + "score": 0.85, + "content": "\\alpha = 0 . 1", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 659, + 459, + 672 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 460, + 659, + 501, + 670 + ], + "score": 0.86, + "content": "\\beta = 0 . 0 1", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 659, + 505, + 672 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 670, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 106, + 670, + 156, + 681 + ], + "score": 0.88, + "content": "\\lambda _ { 1 } = 0 . 0 0 1", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 670, + 161, + 682 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 162, + 670, + 212, + 681 + ], + "score": 0.9, + "content": "\\lambda _ { 2 } = 0 . 0 0 1", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 670, + 231, + 682 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 231, + 670, + 266, + 680 + ], + "score": 0.89, + "content": "\\varepsilon = 0 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 670, + 505, + 682 + ], + "score": 1.0, + "content": ". For recognition, for the comparison with (Busto & Gall,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 680, + 506, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 506, + 694 + ], + "score": 1.0, + "content": "2017) to be fair, we employ a linear SVM classifier in a one-vs-one fashion. Nevertheless, we", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 692, + 505, + 704 + ], + "spans": [ + { + "bbox": [ + 106, + 692, + 245, + 704 + ], + "score": 1.0, + "content": "also report results obtained with a", + "type": "text" + }, + { + "bbox": [ + 245, + 692, + 252, + 702 + ], + "score": 0.54, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 692, + 395, + 704 + ], + "score": 1.0, + "content": "−Nearest-Neighbor classifier (with", + "type": "text" + }, + { + "bbox": [ + 395, + 692, + 421, + 702 + ], + "score": 0.88, + "content": "k = 3", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 692, + 505, + 704 + ], + "score": 1.0, + "content": ") and with our linear", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 702, + 315, + 717 + ], + "spans": [ + { + "bbox": [ + 105, + 702, + 211, + 717 + ], + "score": 1.0, + "content": "classifier with parameters", + "type": "text" + }, + { + "bbox": [ + 211, + 703, + 225, + 713 + ], + "score": 0.4, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 702, + 315, + 717 + ], + "score": 1.0, + "content": "learnt during training.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 30.5 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 114, + 721, + 487, + 732 + ], + "lines": [ + { + "bbox": [ + 119, + 720, + 488, + 733 + ], + "spans": [ + { + "bbox": [ + 119, + 720, + 488, + 733 + ], + "score": 1.0, + "content": "1Among the different variants of ATI (Busto & Gall, 2017), we report the best one in each experiment.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 310, + 761 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 310, + 761 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 108, + 105, + 510, + 276 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 80, + 504, + 101 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 79, + 506, + 92 + ], + "spans": [ + { + "bbox": [ + 105, + 79, + 506, + 92 + ], + "score": 1.0, + "content": "Table 1: Recognition accuracies on the 12 source/target pairs of the BCIS dataset (Tommasi & Tuytelaars,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 90, + 407, + 101 + ], + "spans": [ + { + "bbox": [ + 106, + 90, + 407, + 101 + ], + "score": 1.0, + "content": "2014) using a linear SVM classifier. B: Bing, C: Caltech256, I: ImageNet, S: SUN.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table_body", + "bbox": [ + 108, + 105, + 510, + 276 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 105, + 510, + 276 + ], + "spans": [ + { + "bbox": [ + 108, + 105, + 510, + 276 + ], + "score": 0.98, + "html": "
MethodB→CB→IB→SC→BC→IC→S
TCA (Pan et al.,2011) GFK (Gong et al., 2012)62.8±3.8 56.6± 4.5 66.2 ± 4.0 58.3 ±3.129.6± 4.2 23.8 ±2.038.9±1.9 40.2 ± 1.860.2 ± 1.4 62.2 ± 1.529.7± 1.6 28.5 ± 1.0
SA (Fernando et al.,2013) CORAL (Sun et al., 2016)66.0±3.4 57.8±3.224.3 ± 2.640.3 ± 1.762.5 ± 0.829.0 ± 1.5
ATI (Busto & Gall, 2017)68.8 ±3.3 60.9 ± 2.627.2 ±3.940.7 ± 1.564.0 ± 2.631.4±0.8
AODA (Saito et al., 2018)71.4 ± 2.3 69.0±2.8 76.2 ±1.7 70.9 ±3.237.4± 2.6 57.3 ± 1.145.7 ± 3.0 63.5 ± 2.167.9 ± 4.2 73.5 ± 0.837.5 ± 2.7 60.5±0.8
FRODA D-FRODA73.8±6.1 71.0± 2.0 74.6 ± 5.5 71.4± 2.054.7 ± 2.9 55.4± 2.767.5 ± 1.4 67.6 ± 1.274.5 ± 1.7 75.0±1.861.6±2.2 61.7 ± 2.1
Method TCA (Pan et al.,2011)I→B I→C 40.9±2.9 68.6±1.8I→S 34.5±3.8S→B 19.4 ± 2.1S→C 32.0±3.9S→I Avg. 31.1 ± 4.6 42±3.04
GFK(Gong et al.,2012) SA (Fernando et al.,2013) CORAL (Sun et al.,2016)42.6 ± 2.4 73.3± 3.6 43.1 ± 1.6 72.8 ± 3.1 44.6 ± 2.532.7± 3.6 32.2±3.716.9 ± 1.5 17.5 ± 1.628.6±3.8 26.4± 1.1 29.2 ± 4.2 27.1 ± 1.341.6 ± 2.5 41.8 ± 2.4
ATI (Busto & Gall,2017) AODA (Saito et al., 2018)74.5 ± 3.4 48.8±2.3 77.5 ± 2.2 66.3± 0.9 78.1± 0.935.4 ± 4.4 43.4± 4.818.7 ± 1.2 23.2 ±3.233.6 ± 5.3 31.3 ± 1.3 47.3±2.944.3 ± 2.7 50.2 ± 2.8
FRODA D-FRODA66.0±1.9 79.9 ± 1.7 66.4±1.7 80.5 ± 1.659.4± 1.4 59.2± 2.1 55.7 ± 2.5 59.8 ±2.0 55.5 ± 2.456.5 ± 2.633.0 ±1.1 59.6 ±3.1 63.2 ± 1.3 61.2 ± 1.8 59.4± 1.965.4 ±1.7 65.4± 2.3
", + "type": "table", + "image_path": "8ee7e9346a61f58f347d8877435cc5a9ffe24cc2ce8caac2212b2230efb43686.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 108, + 105, + 510, + 162.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 108, + 162.0, + 510, + 219.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 108, + 219.0, + 510, + 276.0 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 1.75 + }, + { + "type": "text", + "bbox": [ + 106, + 286, + 479, + 298 + ], + "lines": [ + { + "bbox": [ + 107, + 285, + 479, + 299 + ], + "spans": [ + { + "bbox": [ + 107, + 286, + 118, + 296 + ], + "score": 0.84, + "content": "S ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 118, + 285, + 299, + 299 + ], + "score": 1.0, + "content": "-minimization: To minimize equation 7 w.r.t.", + "type": "text" + }, + { + "bbox": [ + 299, + 286, + 311, + 297 + ], + "score": 0.86, + "content": "S ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 285, + 479, + 299 + ], + "score": 1.0, + "content": ", with all other parameters fixed, we write", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5, + "bbox_fs": [ + 107, + 285, + 479, + 299 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 151, + 300, + 457, + 333 + ], + "lines": [ + { + "bbox": [ + 151, + 300, + 457, + 333 + ], + "spans": [ + { + "bbox": [ + 151, + 300, + 457, + 333 + ], + "score": 0.93, + "content": "\\operatorname* { m i n } _ { { \\boldsymbol { S } } ^ { \\prime } } \\quad \\alpha \\| { \\boldsymbol { X } } ^ { \\prime } { \\boldsymbol { s } } - { \\boldsymbol { B } } ^ { \\prime } { \\boldsymbol { S } } ^ { \\prime } \\| _ { F } ^ { 2 } + \\beta \\| { \\boldsymbol { L } } - { \\boldsymbol { W } } ^ { \\prime } { \\boldsymbol { S } } ^ { \\prime } \\| _ { F } ^ { 2 } + \\lambda _ { 2 } \\sum _ { i = 1 } ^ { n _ { s } } \\left( \\| { \\boldsymbol { S } } ^ { \\prime } { \\boldsymbol { s } } ^ { \\prime } \\| + \\| { \\boldsymbol { S } } ^ { \\prime } { \\boldsymbol { u } } \\| \\right) ~ .", + "type": "interline_equation", + "image_path": "83aa450125b6ed8ad171484ca199a1b14a521bc1902386d2cdec09b3ef04749a.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 151, + 300, + 457, + 311.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 151, + 311.0, + 457, + 322.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 151, + 322.0, + 457, + 333.0 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 334, + 505, + 368 + ], + "lines": [ + { + "bbox": [ + 105, + 333, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 505, + 348 + ], + "score": 1.0, + "content": "The first two terms can be grouped into a single squared Frobenius norm, thus resulting in a sparse", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 345, + 505, + 359 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 298, + 359 + ], + "score": 1.0, + "content": "group lasso problem, which, as when updating", + "type": "text" + }, + { + "bbox": [ + 298, + 346, + 307, + 356 + ], + "score": 0.81, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 345, + 505, + 359 + ], + "score": 1.0, + "content": "in FRODA, can be solved via proximal gradient", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 356, + 223, + 369 + ], + "spans": [ + { + "bbox": [ + 106, + 356, + 223, + 369 + ], + "score": 1.0, + "content": "descent (Mairal et al., 2014).", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 333, + 505, + 369 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 372, + 444, + 385 + ], + "lines": [ + { + "bbox": [ + 107, + 369, + 446, + 389 + ], + "spans": [ + { + "bbox": [ + 107, + 373, + 120, + 383 + ], + "score": 0.84, + "content": "B ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 120, + 369, + 271, + 389 + ], + "score": 1.0, + "content": "-minimization: Given the coefficients", + "type": "text" + }, + { + "bbox": [ + 271, + 373, + 287, + 383 + ], + "score": 0.89, + "content": "S ^ { \\prime ^ { v } }", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 369, + 305, + 389 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 306, + 373, + 322, + 383 + ], + "score": 0.89, + "content": "S ^ { \\prime } { } ^ { u }", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 369, + 385, + 389 + ], + "score": 1.0, + "content": ", we can update", + "type": "text" + }, + { + "bbox": [ + 385, + 374, + 399, + 383 + ], + "score": 0.87, + "content": "U ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 369, + 446, + 389 + ], + "score": 1.0, + "content": "by solving", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12, + "bbox_fs": [ + 107, + 369, + 446, + 389 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 210, + 387, + 399, + 423 + ], + "lines": [ + { + "bbox": [ + 210, + 387, + 399, + 423 + ], + "spans": [ + { + "bbox": [ + 210, + 387, + 399, + 423 + ], + "score": 0.94, + "content": "\\operatorname* { m i n } _ { U ^ { \\prime } } \\quad \\| A - U ^ { \\prime } S ^ { \\prime ^ { \\# } } \\| _ { F } ^ { 2 } \\quad s . t . \\sum _ { j = 1 } ^ { d } \\| U ^ { \\prime } { } _ { j } \\| ^ { 2 } \\leq 1 ,", + "type": "interline_equation", + "image_path": "e064d3974e5e9c0fb0c605063d71cb08f06bd13e992196deb75d0d3a0109b414.jpg" + } + ] + } + ], + "index": 13.5, + "virtual_lines": [ + { + "bbox": [ + 210, + 387, + 399, + 405.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 210, + 405.0, + 399, + 423.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 425, + 281, + 438 + ], + "lines": [ + { + "bbox": [ + 105, + 422, + 283, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 422, + 127, + 442 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 127, + 425, + 204, + 438 + ], + "score": 0.92, + "content": "A = { X ^ { \\prime } } _ { s } - V { S ^ { \\prime } } ^ { v }", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 422, + 225, + 442 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 225, + 426, + 235, + 436 + ], + "score": 0.82, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 422, + 283, + 442 + ], + "score": 1.0, + "content": "by solving", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15, + "bbox_fs": [ + 105, + 422, + 283, + 442 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 216, + 440, + 393, + 475 + ], + "lines": [ + { + "bbox": [ + 216, + 440, + 393, + 475 + ], + "spans": [ + { + "bbox": [ + 216, + 440, + 393, + 475 + ], + "score": 0.94, + "content": "\\operatorname* { m i n } _ { V } \\quad \\| A ^ { \\prime } - V C \\| _ { F } ^ { 2 } \\quad s . t . \\sum _ { j = 1 } ^ { d } \\| V _ { j } \\| ^ { 2 } \\leq 1 ,", + "type": "interline_equation", + "image_path": "e11cb5d6fcdf5370381a318c998557aef3776053c73c21d1ac6ad328e9b12fe8.jpg" + } + ] + } + ], + "index": 16.5, + "virtual_lines": [ + { + "bbox": [ + 216, + 440, + 393, + 457.5 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 216, + 457.5, + 393, + 475.0 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 478, + 505, + 515 + ], + "lines": [ + { + "bbox": [ + 104, + 468, + 514, + 511 + ], + "spans": [ + { + "bbox": [ + 104, + 479, + 127, + 500 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 127, + 478, + 250, + 504 + ], + "score": 0.93, + "content": "A ^ { \\prime } = \\binom { X _ { t } - U T ^ { u } } { \\sqrt { \\alpha } ( X ^ { \\prime } { } _ { s } - U ^ { \\prime } S ^ { \\prime } { } ^ { u } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 468, + 269, + 511 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 270, + 478, + 338, + 505 + ], + "score": 0.94, + "content": "C = \\left( \\underset { \\sqrt { \\alpha } S ^ { \\prime } } { \\mathbf { \\nabla } } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 468, + 514, + 511 + ], + "score": 1.0, + "content": ". As in FRODA, these two sub-problems", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 500, + 436, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 436, + 516 + ], + "score": 1.0, + "content": "can be solved efficiently using the Lagrange dual formulation of Lee et al. (2007).", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 104, + 468, + 514, + 516 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 527, + 201, + 540 + ], + "lines": [ + { + "bbox": [ + 104, + 525, + 202, + 542 + ], + "spans": [ + { + "bbox": [ + 104, + 525, + 202, + 542 + ], + "score": 1.0, + "content": "4 EXPERIMENTS", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 106, + 545, + 505, + 601 + ], + "lines": [ + { + "bbox": [ + 106, + 546, + 505, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 505, + 558 + ], + "score": 1.0, + "content": "We evaluate our approach on the task of open-set visual domain adaptation using two benchmark", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 556, + 504, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 556, + 504, + 569 + ], + "score": 1.0, + "content": "datasets, and compare its performance against the state-of-the-art open-set domain adaptation meth-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 568, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 106, + 568, + 505, + 579 + ], + "score": 1.0, + "content": "ods on each dataset.1 Note that we also report the results of the methods used as baselines in (Busto", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 578, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 106, + 578, + 246, + 591 + ], + "score": 1.0, + "content": "& Gall, 2017). For a dataset with", + "type": "text" + }, + { + "bbox": [ + 246, + 579, + 256, + 588 + ], + "score": 0.8, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 578, + 428, + 591 + ], + "score": 1.0, + "content": "source classes, we report the accuracy on", + "type": "text" + }, + { + "bbox": [ + 428, + 579, + 455, + 589 + ], + "score": 0.9, + "content": "C + 1", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 578, + 505, + 591 + ], + "score": 1.0, + "content": "classes, the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 588, + 311, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 311, + 603 + ], + "score": 1.0, + "content": "additional one corresponding to the unknown case.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 546, + 505, + 603 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 603, + 505, + 714 + ], + "lines": [ + { + "bbox": [ + 105, + 603, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 505, + 617 + ], + "score": 1.0, + "content": "Implementation details. Following Busto & Gall (2017), we represent the source and target sam-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 615, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 220, + 627 + ], + "score": 1.0, + "content": "ples with 4096-dimensional", + "type": "text" + }, + { + "bbox": [ + 221, + 615, + 261, + 626 + ], + "score": 0.92, + "content": "D e C A F _ { 7 }", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 615, + 505, + 627 + ], + "score": 1.0, + "content": "features (Donahue et al., 2013) and first reduce their dimen-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 626, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 505, + 639 + ], + "score": 1.0, + "content": "sionality by performing PCA jointly on the source and target data and keeping the components", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 636, + 506, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 145, + 650 + ], + "score": 1.0, + "content": "encoding", + "type": "text" + }, + { + "bbox": [ + 146, + 637, + 166, + 648 + ], + "score": 0.86, + "content": "9 9 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 636, + 406, + 650 + ], + "score": 1.0, + "content": "of the data variance. To then determine the dimensionality", + "type": "text" + }, + { + "bbox": [ + 407, + 637, + 414, + 647 + ], + "score": 0.74, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 636, + 506, + 650 + ], + "score": 1.0, + "content": "of our shared and pri-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 648, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 505, + 660 + ], + "score": 1.0, + "content": "vate subspaces, we make use of the subspace disagreement measure of (Gong et al., 2012). For all", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 659, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 417, + 672 + ], + "score": 1.0, + "content": "our experiments, the hyperparameters of our approach were set as follows:", + "type": "text" + }, + { + "bbox": [ + 417, + 659, + 454, + 669 + ], + "score": 0.85, + "content": "\\alpha = 0 . 1", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 659, + 459, + 672 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 460, + 659, + 501, + 670 + ], + "score": 0.86, + "content": "\\beta = 0 . 0 1", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 659, + 505, + 672 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 670, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 106, + 670, + 156, + 681 + ], + "score": 0.88, + "content": "\\lambda _ { 1 } = 0 . 0 0 1", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 670, + 161, + 682 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 162, + 670, + 212, + 681 + ], + "score": 0.9, + "content": "\\lambda _ { 2 } = 0 . 0 0 1", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 670, + 231, + 682 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 231, + 670, + 266, + 680 + ], + "score": 0.89, + "content": "\\varepsilon = 0 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 670, + 505, + 682 + ], + "score": 1.0, + "content": ". For recognition, for the comparison with (Busto & Gall,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 680, + 506, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 506, + 694 + ], + "score": 1.0, + "content": "2017) to be fair, we employ a linear SVM classifier in a one-vs-one fashion. Nevertheless, we", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 692, + 505, + 704 + ], + "spans": [ + { + "bbox": [ + 106, + 692, + 245, + 704 + ], + "score": 1.0, + "content": "also report results obtained with a", + "type": "text" + }, + { + "bbox": [ + 245, + 692, + 252, + 702 + ], + "score": 0.54, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 692, + 395, + 704 + ], + "score": 1.0, + "content": "−Nearest-Neighbor classifier (with", + "type": "text" + }, + { + "bbox": [ + 395, + 692, + 421, + 702 + ], + "score": 0.88, + "content": "k = 3", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 692, + 505, + 704 + ], + "score": 1.0, + "content": ") and with our linear", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 702, + 315, + 717 + ], + "spans": [ + { + "bbox": [ + 105, + 702, + 211, + 717 + ], + "score": 1.0, + "content": "classifier with parameters", + "type": "text" + }, + { + "bbox": [ + 211, + 703, + 225, + 713 + ], + "score": 0.4, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 702, + 315, + 717 + ], + "score": 1.0, + "content": "learnt during training.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 603, + 506, + 717 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 108, + 114, + 509, + 304 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 80, + 506, + 110 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 79, + 505, + 92 + ], + "spans": [ + { + "bbox": [ + 105, + 79, + 505, + 92 + ], + "score": 1.0, + "content": "Table 2: Recognition accuracies of variants of our approach using linear SVM, nearest neighbor (NN) and our", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 89, + 506, + 103 + ], + "spans": [ + { + "bbox": [ + 105, + 89, + 164, + 103 + ], + "score": 1.0, + "content": "linear classifier", + "type": "text" + }, + { + "bbox": [ + 164, + 91, + 181, + 100 + ], + "score": 0.44, + "content": "( W )", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 89, + 506, + 103 + ], + "score": 1.0, + "content": "on the 12 source/target pairs of the BCIS dataset (Tommasi & Tuytelaars, 2014). B: Bing,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 100, + 244, + 111 + ], + "spans": [ + { + "bbox": [ + 106, + 100, + 244, + 111 + ], + "score": 1.0, + "content": "C: Caltech256, I: ImageNet, S: SUN.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "table_body", + "bbox": [ + 108, + 114, + 509, + 304 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 114, + 509, + 304 + ], + "spans": [ + { + "bbox": [ + 108, + 114, + 509, + 304 + ], + "score": 0.754, + "html": "
MethodB→CB→IB→SC→BC→IC→S
FRODA-SVM73.8±6.171.0±2.0 64.1 ± 2.454.7±2.9 54.4 ±3.567.5 ± 1.4 65.1 ± 2.974.5 ± 1.7 72.9 ±1.761.6± 2.2
FRODA-NN D-FRODA-SVM D-FRODA-W67.7 ±2.8 74.6± 5.5 61.2 ± 1.271.4± 2.0 59.3 ± 1.155.4± 2.7 53.3 ± 3.067.6±1.2 63.1±0.975.0±1.8 66.2 ± 1.760.5 ± 2.0 61.7± 2.1 60.2 ± 2.1
D-FRODA-NN D-FRODA-U-SVM D-FRODA-U-W67.7 ± 3.3 71.9±3.8 55.9 ± 3.663.3 ± 2.8 69.2± 2.7 55.5± 4.554.0 ± 3.5 56.7± 3.3 41.7 ± 4.865.4± 2.8 64.8±1.8 52.6 ± 4.773.4 ± 1.5 72.8±2.7 61.4 ± 3.660.3 ± 2.4 59.4± 2.3 49.5 ± 4.1
D-FRODA-U-NN57.6 ± 9.152.2 ± 5.047.5 ± 6.551.8±5.764.0±5.757.7 ± 5.6
MethodI→BI→CI→SS→BS→CS→IAvg.
FRODA-SVM FRODA-NN66.0±1.9 60.9 ± 3.779.9 ± 1.7 77.7 ± 2.859.2 ± 2.1 58.0± 2.255.7± 2.5 53.4 ± 2.261.2 ± 1.859.4 ± 1.965.4
D-FRODA-SVM66.4±1.780.5±1.659.8± 2.055.5± 2.461.2 ± 1.4 61.2 ± 1.958.1 ± 1.5 59.6±2.262.8 65.7
D-FRODA-W62.2 ± 1.670.5 ± 2.558.1 ± 1.756.4 ±1.958.9 ± 1.558.5±0.760.7
D-FRODA-NN D-FRODA-U-SVM60.9 ± 4.178.7± 2.857.7± 2.053.0±2.361.2 ±1.257.9 ± 1.662.8
66.0± 1.276.8± 1.957.2 ± 4.556.3±1.961.8± 3.059.9±1.764.4
D-FRODA-U-W54.6 ± 4.466.2 ±3.843.9 ± 5.447.6 ± 3.851.3 ±3.7
53.5 ± 4.952.8
58.4±6.170.9 ± 4.552.8 ±9.055.4 ± 2.0
D-FRODA-U-NN61.5 ± 2.060.1 ±1.757.5
", + "type": "table", + "image_path": "a518c73202ee86aca28c5911a774c9336b156fd29f7e95cc8643bf18110f8ee9.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 108, + 114, + 509, + 177.33333333333334 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 108, + 177.33333333333334, + 509, + 240.66666666666669 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 108, + 240.66666666666669, + 509, + 304.0 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 106, + 319, + 505, + 451 + ], + "lines": [ + { + "bbox": [ + 105, + 317, + 505, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 505, + 333 + ], + "score": 1.0, + "content": "Results on the dense cross-dataset benchmark. We first evaluate our approach on the challeng-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 330, + 506, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 330, + 506, + 343 + ], + "score": 1.0, + "content": "ing cross-dataset benchmark of Tommasi & Tuytelaars (2014). This dataset was built using images", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 341, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 106, + 341, + 505, + 353 + ], + "score": 1.0, + "content": "depicting 40 object categories and coming from four datasets, namely Bing (B), Caltech256 (C),", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 352, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 505, + 364 + ], + "score": 1.0, + "content": "ImageNet (I) and SUN (S), hence referred to as BCIS. Following Busto & Gall (2017), we con-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 363, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 106, + 363, + 505, + 374 + ], + "score": 1.0, + "content": "sider the samples from the first 10 classes as known instances, while the samples with class labels", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 107, + 373, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 107, + 374, + 166, + 385 + ], + "score": 0.77, + "content": "1 1 , 1 2 , \\cdots , 2 5", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 373, + 184, + 386 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 184, + 374, + 244, + 385 + ], + "score": 0.64, + "content": "2 6 , 2 7 , \\cdots , 4 0", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 373, + 505, + 386 + ], + "score": 1.0, + "content": "are taken to be the unknown samples in the source and target do-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 385, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 505, + 397 + ], + "score": 1.0, + "content": "mains, respectively. We follow the unsupervised protocol of Tommasi & Tuytelaars (2014), which", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 396, + 506, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 506, + 408 + ], + "score": 1.0, + "content": "relies on 50 source samples per class and 30 target images per class, except when the target data", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 104, + 406, + 505, + 420 + ], + "spans": [ + { + "bbox": [ + 104, + 406, + 505, + 420 + ], + "score": 1.0, + "content": "is coming from SUN, in which case only 20 images per class are employed. Note that only the", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 416, + 505, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 418, + 147, + 429 + ], + "score": 0.9, + "content": "D e C A F _ { 7 }", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 416, + 505, + 430 + ], + "score": 1.0, + "content": "features are publicly available. To nonetheless evaluate the AODA method of Saito et al.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 429, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 429, + 289, + 441 + ], + "score": 1.0, + "content": "(2018), we made use of a network taking the", + "type": "text" + }, + { + "bbox": [ + 289, + 429, + 330, + 440 + ], + "score": 0.91, + "content": "D e C A F _ { 7 }", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 429, + 505, + 441 + ], + "score": 1.0, + "content": "features as input and processing them with", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 439, + 493, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 493, + 452 + ], + "score": 1.0, + "content": "two fully-connected layers, with 1024 and 128 units, respectively, and a final classification layer.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 106, + 456, + 505, + 534 + ], + "lines": [ + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "score": 1.0, + "content": "In Table 1, we compare the results of our methods with those of the baselines on all 12 domain pairs", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 467, + 504, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 504, + 480 + ], + "score": 1.0, + "content": "of this dataset. Note that our algorithms (both with and without the discriminative term) outperform", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 104, + 477, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 104, + 477, + 505, + 492 + ], + "score": 1.0, + "content": "all the baselines, and in particular the state-of-the-art one of Busto & Gall (2017) by a large margin.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "score": 1.0, + "content": "For instance, the margin exceeds 32, resp. 26, percentage points when going from SUN to Bing and", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 104, + 499, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 104, + 499, + 505, + 514 + ], + "score": 1.0, + "content": "ImageNet, respectively. This, we believe, clearly evidences the benefits of our factorized represen-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 511, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 505, + 525 + ], + "score": 1.0, + "content": "tations, which allow us to separate the unknown target samples from the ones coming from known", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 523, + 375, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 375, + 534 + ], + "score": 1.0, + "content": "classes, thus yielding a better representation for the known classes.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 540, + 297, + 704 + ], + "lines": [ + { + "bbox": [ + 106, + 539, + 297, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 297, + 551 + ], + "score": 1.0, + "content": "In Table 2, we compare different versions of our", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 550, + 297, + 562 + ], + "spans": [ + { + "bbox": [ + 106, + 550, + 297, + 562 + ], + "score": 1.0, + "content": "method, corresponding to using different clas-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 561, + 297, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 297, + 573 + ], + "score": 1.0, + "content": "sifiers and to using additional unknown source", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 572, + 297, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 297, + 583 + ], + "score": 1.0, + "content": "data. Note that the linear SVM classifier, when", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 583, + 297, + 594 + ], + "spans": [ + { + "bbox": [ + 106, + 583, + 297, + 594 + ], + "score": 1.0, + "content": "used with our framework, tends to perform the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 594, + 297, + 605 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 297, + 605 + ], + "score": 1.0, + "content": "best, followed by the NN one and finally the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 605, + 297, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 297, + 617 + ], + "score": 1.0, + "content": "learnt linear classifier. This, we believe, can be", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 617, + 297, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 617, + 297, + 627 + ], + "score": 1.0, + "content": "explained by the fact that, while the linear clas-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 626, + 297, + 639 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 297, + 639 + ], + "score": 1.0, + "content": "sifier helps to learn a more discriminative rep-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 638, + 297, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 297, + 650 + ], + "score": 1.0, + "content": "resentation, it remains less powerful than the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 648, + 297, + 661 + ], + "spans": [ + { + "bbox": [ + 106, + 648, + 297, + 661 + ], + "score": 1.0, + "content": "other two classifiers to label the target sam-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 659, + 298, + 672 + ], + "spans": [ + { + "bbox": [ + 106, + 659, + 298, + 672 + ], + "score": 1.0, + "content": "ples. Note also that the use of unknown source", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 671, + 297, + 683 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 297, + 683 + ], + "score": 1.0, + "content": "data does not consistently help in our frame-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 681, + 296, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 681, + 296, + 693 + ], + "score": 1.0, + "content": "work. Nevertheless, the corresponding results", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 692, + 293, + 705 + ], + "spans": [ + { + "bbox": [ + 106, + 692, + 293, + 705 + ], + "score": 1.0, + "content": "still outperform those of Busto & Gall (2017).", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 106, + 710, + 506, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 297, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 297, + 723 + ], + "score": 1.0, + "content": "We further evaluate the robustness of our ap-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 241, + 733 + ], + "score": 1.0, + "content": "proach to the choice of threshold", + "type": "text" + }, + { + "bbox": [ + 241, + 722, + 247, + 730 + ], + "score": 0.72, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "to separate the target samples from known/unknown classes. In", + "type": "text" + } + ], + "index": 54 + } + ], + "index": 47.0 + }, + { + "type": "image", + "bbox": [ + 306, + 540, + 501, + 696 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 306, + 540, + 501, + 696 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 306, + 540, + 501, + 696 + ], + "spans": [ + { + "bbox": [ + 306, + 540, + 501, + 696 + ], + "score": 0.966, + "type": "image", + "image_path": "f6e8307a91cba11a9779b2312f183596f19aa74ded692c38ca1f932f398ec5fb.jpg" + } + ] + } + ], + "index": 46.5, + "virtual_lines": [ + { + "bbox": [ + 306, + 540, + 501, + 553.0 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 306, + 553.0, + 501, + 566.0 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 306, + 566.0, + 501, + 579.0 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 306, + 579.0, + 501, + 592.0 + ], + "spans": [], + "index": 44 + }, + { + "bbox": [ + 306, + 592.0, + 501, + 605.0 + ], + "spans": [], + "index": 45 + }, + { + "bbox": [ + 306, + 605.0, + 501, + 618.0 + ], + "spans": [], + "index": 46 + }, + { + "bbox": [ + 306, + 618.0, + 501, + 631.0 + ], + "spans": [], + "index": 47 + }, + { + "bbox": [ + 306, + 631.0, + 501, + 644.0 + ], + "spans": [], + "index": 48 + }, + { + "bbox": [ + 306, + 644.0, + 501, + 657.0 + ], + "spans": [], + "index": 49 + }, + { + "bbox": [ + 306, + 657.0, + 501, + 670.0 + ], + "spans": [], + "index": 50 + }, + { + "bbox": [ + 306, + 670.0, + 501, + 683.0 + ], + "spans": [], + "index": 51 + }, + { + "bbox": [ + 306, + 683.0, + 501, + 696.0 + ], + "spans": [], + "index": 52 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 356, + 699, + 452, + 710 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 355, + 698, + 452, + 712 + ], + "spans": [ + { + "bbox": [ + 355, + 698, + 446, + 712 + ], + "score": 1.0, + "content": "Figure 1: Sensitivity to", + "type": "text" + }, + { + "bbox": [ + 446, + 702, + 452, + 709 + ], + "score": 0.52, + "content": "\\varepsilon", + "type": "inline_equation" + } + ], + "index": 53 + } + ], + "index": 53 + } + ], + "index": 49.75 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 108, + 114, + 509, + 304 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 80, + 506, + 110 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 79, + 505, + 92 + ], + "spans": [ + { + "bbox": [ + 105, + 79, + 505, + 92 + ], + "score": 1.0, + "content": "Table 2: Recognition accuracies of variants of our approach using linear SVM, nearest neighbor (NN) and our", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 89, + 506, + 103 + ], + "spans": [ + { + "bbox": [ + 105, + 89, + 164, + 103 + ], + "score": 1.0, + "content": "linear classifier", + "type": "text" + }, + { + "bbox": [ + 164, + 91, + 181, + 100 + ], + "score": 0.44, + "content": "( W )", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 89, + 506, + 103 + ], + "score": 1.0, + "content": "on the 12 source/target pairs of the BCIS dataset (Tommasi & Tuytelaars, 2014). B: Bing,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 100, + 244, + 111 + ], + "spans": [ + { + "bbox": [ + 106, + 100, + 244, + 111 + ], + "score": 1.0, + "content": "C: Caltech256, I: ImageNet, S: SUN.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "table_body", + "bbox": [ + 108, + 114, + 509, + 304 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 114, + 509, + 304 + ], + "spans": [ + { + "bbox": [ + 108, + 114, + 509, + 304 + ], + "score": 0.754, + "html": "
MethodB→CB→IB→SC→BC→IC→S
FRODA-SVM73.8±6.171.0±2.0 64.1 ± 2.454.7±2.9 54.4 ±3.567.5 ± 1.4 65.1 ± 2.974.5 ± 1.7 72.9 ±1.761.6± 2.2
FRODA-NN D-FRODA-SVM D-FRODA-W67.7 ±2.8 74.6± 5.5 61.2 ± 1.271.4± 2.0 59.3 ± 1.155.4± 2.7 53.3 ± 3.067.6±1.2 63.1±0.975.0±1.8 66.2 ± 1.760.5 ± 2.0 61.7± 2.1 60.2 ± 2.1
D-FRODA-NN D-FRODA-U-SVM D-FRODA-U-W67.7 ± 3.3 71.9±3.8 55.9 ± 3.663.3 ± 2.8 69.2± 2.7 55.5± 4.554.0 ± 3.5 56.7± 3.3 41.7 ± 4.865.4± 2.8 64.8±1.8 52.6 ± 4.773.4 ± 1.5 72.8±2.7 61.4 ± 3.660.3 ± 2.4 59.4± 2.3 49.5 ± 4.1
D-FRODA-U-NN57.6 ± 9.152.2 ± 5.047.5 ± 6.551.8±5.764.0±5.757.7 ± 5.6
MethodI→BI→CI→SS→BS→CS→IAvg.
FRODA-SVM FRODA-NN66.0±1.9 60.9 ± 3.779.9 ± 1.7 77.7 ± 2.859.2 ± 2.1 58.0± 2.255.7± 2.5 53.4 ± 2.261.2 ± 1.859.4 ± 1.965.4
D-FRODA-SVM66.4±1.780.5±1.659.8± 2.055.5± 2.461.2 ± 1.4 61.2 ± 1.958.1 ± 1.5 59.6±2.262.8 65.7
D-FRODA-W62.2 ± 1.670.5 ± 2.558.1 ± 1.756.4 ±1.958.9 ± 1.558.5±0.760.7
D-FRODA-NN D-FRODA-U-SVM60.9 ± 4.178.7± 2.857.7± 2.053.0±2.361.2 ±1.257.9 ± 1.662.8
66.0± 1.276.8± 1.957.2 ± 4.556.3±1.961.8± 3.059.9±1.764.4
D-FRODA-U-W54.6 ± 4.466.2 ±3.843.9 ± 5.447.6 ± 3.851.3 ±3.7
53.5 ± 4.952.8
58.4±6.170.9 ± 4.552.8 ±9.055.4 ± 2.0
D-FRODA-U-NN61.5 ± 2.060.1 ±1.757.5
", + "type": "table", + "image_path": "a518c73202ee86aca28c5911a774c9336b156fd29f7e95cc8643bf18110f8ee9.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 108, + 114, + 509, + 177.33333333333334 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 108, + 177.33333333333334, + 509, + 240.66666666666669 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 108, + 240.66666666666669, + 509, + 304.0 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 106, + 319, + 505, + 451 + ], + "lines": [ + { + "bbox": [ + 105, + 317, + 505, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 505, + 333 + ], + "score": 1.0, + "content": "Results on the dense cross-dataset benchmark. We first evaluate our approach on the challeng-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 330, + 506, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 330, + 506, + 343 + ], + "score": 1.0, + "content": "ing cross-dataset benchmark of Tommasi & Tuytelaars (2014). This dataset was built using images", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 341, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 106, + 341, + 505, + 353 + ], + "score": 1.0, + "content": "depicting 40 object categories and coming from four datasets, namely Bing (B), Caltech256 (C),", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 352, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 505, + 364 + ], + "score": 1.0, + "content": "ImageNet (I) and SUN (S), hence referred to as BCIS. Following Busto & Gall (2017), we con-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 363, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 106, + 363, + 505, + 374 + ], + "score": 1.0, + "content": "sider the samples from the first 10 classes as known instances, while the samples with class labels", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 107, + 373, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 107, + 374, + 166, + 385 + ], + "score": 0.77, + "content": "1 1 , 1 2 , \\cdots , 2 5", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 373, + 184, + 386 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 184, + 374, + 244, + 385 + ], + "score": 0.64, + "content": "2 6 , 2 7 , \\cdots , 4 0", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 373, + 505, + 386 + ], + "score": 1.0, + "content": "are taken to be the unknown samples in the source and target do-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 385, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 505, + 397 + ], + "score": 1.0, + "content": "mains, respectively. We follow the unsupervised protocol of Tommasi & Tuytelaars (2014), which", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 396, + 506, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 506, + 408 + ], + "score": 1.0, + "content": "relies on 50 source samples per class and 30 target images per class, except when the target data", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 104, + 406, + 505, + 420 + ], + "spans": [ + { + "bbox": [ + 104, + 406, + 505, + 420 + ], + "score": 1.0, + "content": "is coming from SUN, in which case only 20 images per class are employed. Note that only the", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 416, + 505, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 418, + 147, + 429 + ], + "score": 0.9, + "content": "D e C A F _ { 7 }", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 416, + 505, + 430 + ], + "score": 1.0, + "content": "features are publicly available. To nonetheless evaluate the AODA method of Saito et al.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 429, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 429, + 289, + 441 + ], + "score": 1.0, + "content": "(2018), we made use of a network taking the", + "type": "text" + }, + { + "bbox": [ + 289, + 429, + 330, + 440 + ], + "score": 0.91, + "content": "D e C A F _ { 7 }", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 429, + 505, + 441 + ], + "score": 1.0, + "content": "features as input and processing them with", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 439, + 493, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 493, + 452 + ], + "score": 1.0, + "content": "two fully-connected layers, with 1024 and 128 units, respectively, and a final classification layer.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 11.5, + "bbox_fs": [ + 104, + 317, + 506, + 452 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 456, + 505, + 534 + ], + "lines": [ + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "score": 1.0, + "content": "In Table 1, we compare the results of our methods with those of the baselines on all 12 domain pairs", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 467, + 504, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 504, + 480 + ], + "score": 1.0, + "content": "of this dataset. Note that our algorithms (both with and without the discriminative term) outperform", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 104, + 477, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 104, + 477, + 505, + 492 + ], + "score": 1.0, + "content": "all the baselines, and in particular the state-of-the-art one of Busto & Gall (2017) by a large margin.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "score": 1.0, + "content": "For instance, the margin exceeds 32, resp. 26, percentage points when going from SUN to Bing and", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 104, + 499, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 104, + 499, + 505, + 514 + ], + "score": 1.0, + "content": "ImageNet, respectively. This, we believe, clearly evidences the benefits of our factorized represen-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 511, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 505, + 525 + ], + "score": 1.0, + "content": "tations, which allow us to separate the unknown target samples from the ones coming from known", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 523, + 375, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 375, + 534 + ], + "score": 1.0, + "content": "classes, thus yielding a better representation for the known classes.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21, + "bbox_fs": [ + 104, + 456, + 505, + 534 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 540, + 297, + 704 + ], + "lines": [ + { + "bbox": [ + 106, + 539, + 297, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 297, + 551 + ], + "score": 1.0, + "content": "In Table 2, we compare different versions of our", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 550, + 297, + 562 + ], + "spans": [ + { + "bbox": [ + 106, + 550, + 297, + 562 + ], + "score": 1.0, + "content": "method, corresponding to using different clas-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 561, + 297, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 297, + 573 + ], + "score": 1.0, + "content": "sifiers and to using additional unknown source", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 572, + 297, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 297, + 583 + ], + "score": 1.0, + "content": "data. Note that the linear SVM classifier, when", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 583, + 297, + 594 + ], + "spans": [ + { + "bbox": [ + 106, + 583, + 297, + 594 + ], + "score": 1.0, + "content": "used with our framework, tends to perform the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 594, + 297, + 605 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 297, + 605 + ], + "score": 1.0, + "content": "best, followed by the NN one and finally the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 605, + 297, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 297, + 617 + ], + "score": 1.0, + "content": "learnt linear classifier. This, we believe, can be", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 617, + 297, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 617, + 297, + 627 + ], + "score": 1.0, + "content": "explained by the fact that, while the linear clas-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 626, + 297, + 639 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 297, + 639 + ], + "score": 1.0, + "content": "sifier helps to learn a more discriminative rep-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 638, + 297, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 297, + 650 + ], + "score": 1.0, + "content": "resentation, it remains less powerful than the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 648, + 297, + 661 + ], + "spans": [ + { + "bbox": [ + 106, + 648, + 297, + 661 + ], + "score": 1.0, + "content": "other two classifiers to label the target sam-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 659, + 298, + 672 + ], + "spans": [ + { + "bbox": [ + 106, + 659, + 298, + 672 + ], + "score": 1.0, + "content": "ples. Note also that the use of unknown source", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 671, + 297, + 683 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 297, + 683 + ], + "score": 1.0, + "content": "data does not consistently help in our frame-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 681, + 296, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 681, + 296, + 693 + ], + "score": 1.0, + "content": "work. Nevertheless, the corresponding results", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 692, + 293, + 705 + ], + "spans": [ + { + "bbox": [ + 106, + 692, + 293, + 705 + ], + "score": 1.0, + "content": "still outperform those of Busto & Gall (2017).", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 32, + "bbox_fs": [ + 106, + 539, + 298, + 705 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 710, + 506, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 297, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 297, + 723 + ], + "score": 1.0, + "content": "We further evaluate the robustness of our ap-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 241, + 733 + ], + "score": 1.0, + "content": "proach to the choice of threshold", + "type": "text" + }, + { + "bbox": [ + 241, + 722, + 247, + 730 + ], + "score": 0.72, + "content": "\\varepsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "to separate the target samples from known/unknown classes. In", + "type": "text" + } + ], + "index": 54 + } + ], + "index": 47.0, + "bbox_fs": [ + 105, + 708, + 505, + 733 + ] + }, + { + "type": "image", + "bbox": [ + 306, + 540, + 501, + 696 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 306, + 540, + 501, + 696 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 306, + 540, + 501, + 696 + ], + "spans": [ + { + "bbox": [ + 306, + 540, + 501, + 696 + ], + "score": 0.966, + "type": "image", + "image_path": "f6e8307a91cba11a9779b2312f183596f19aa74ded692c38ca1f932f398ec5fb.jpg" + } + ] + } + ], + "index": 46.5, + "virtual_lines": [ + { + "bbox": [ + 306, + 540, + 501, + 553.0 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 306, + 553.0, + 501, + 566.0 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 306, + 566.0, + 501, + 579.0 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 306, + 579.0, + 501, + 592.0 + ], + "spans": [], + "index": 44 + }, + { + "bbox": [ + 306, + 592.0, + 501, + 605.0 + ], + "spans": [], + "index": 45 + }, + { + "bbox": [ + 306, + 605.0, + 501, + 618.0 + ], + "spans": [], + "index": 46 + }, + { + "bbox": [ + 306, + 618.0, + 501, + 631.0 + ], + "spans": [], + "index": 47 + }, + { + "bbox": [ + 306, + 631.0, + 501, + 644.0 + ], + "spans": [], + "index": 48 + }, + { + "bbox": [ + 306, + 644.0, + 501, + 657.0 + ], + "spans": [], + "index": 49 + }, + { + "bbox": [ + 306, + 657.0, + 501, + 670.0 + ], + "spans": [], + "index": 50 + }, + { + "bbox": [ + 306, + 670.0, + 501, + 683.0 + ], + "spans": [], + "index": 51 + }, + { + "bbox": [ + 306, + 683.0, + 501, + 696.0 + ], + "spans": [], + "index": 52 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 356, + 699, + 452, + 710 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 355, + 698, + 452, + 712 + ], + "spans": [ + { + "bbox": [ + 355, + 698, + 446, + 712 + ], + "score": 1.0, + "content": "Figure 1: Sensitivity to", + "type": "text" + }, + { + "bbox": [ + 446, + 702, + 452, + 709 + ], + "score": 0.52, + "content": "\\varepsilon", + "type": "inline_equation" + } + ], + "index": 53 + } + ], + "index": 53 + } + ], + "index": 49.75 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 108, + 105, + 505, + 218 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 81, + 504, + 100 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 78, + 506, + 93 + ], + "spans": [ + { + "bbox": [ + 105, + 78, + 506, + 93 + ], + "score": 1.0, + "content": "Table 3: Recognition accuracies on the 6 source/target pairs of the Office dataset (Saenko et al., 2010) using a", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 91, + 321, + 100 + ], + "spans": [ + { + "bbox": [ + 106, + 91, + 321, + 100 + ], + "score": 1.0, + "content": "linear SVM classifier. A: Amazon, W: Webcam, D: DSLR.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table_body", + "bbox": [ + 108, + 105, + 505, + 218 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 105, + 505, + 218 + ], + "spans": [ + { + "bbox": [ + 108, + 105, + 505, + 218 + ], + "score": 0.98, + "html": "
MethodA→D A→WW→AW→DD→AD →WAvg.
LSVM72.657.549.298.845.188.568.6
DAN (Long et al.,2016a)77.672.560.898.35788.475.8
RTN (Long et al., 2016b)76.67362.498.857.28976.2
BP(Ganin & Lempitsky,2014)78.375.96498.757.689.877.4
ADDA (Tzeng et al.,2017)52.558.354.189.145.379.163.1
DSN (Bousmalis et al.,2016)58.357.255.179.358.170.263.0
ATI (Busto & Gall,2017)79.878.476.798.871.394.483.2
AODA (Saito et al.,2018)76.674.981.296.962.394.681.1
FRODA88.078.776.598.073.794.684.9
D-FRODA87.478.177.198.573.694.484.9
", + "type": "table", + "image_path": "6af5dcaeacd2773b50d2e1e0ede51fb7a086b5fe46272a7400190bc2ae5d4a5d.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 108, + 105, + 505, + 142.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 108, + 142.66666666666666, + 505, + 180.33333333333331 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 108, + 180.33333333333331, + 505, + 217.99999999999997 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 1.75 + }, + { + "type": "table", + "bbox": [ + 110, + 260, + 503, + 367 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 235, + 505, + 256 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 235, + 505, + 246 + ], + "spans": [ + { + "bbox": [ + 106, + 235, + 505, + 246 + ], + "score": 1.0, + "content": "Table 4: Recognition accuracies of variants of our approaches using a linear SVM, nearest neighbor (NN) and", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 245, + 449, + 257 + ], + "spans": [ + { + "bbox": [ + 106, + 245, + 178, + 257 + ], + "score": 1.0, + "content": "our linear classifier", + "type": "text" + }, + { + "bbox": [ + 179, + 245, + 196, + 255 + ], + "score": 0.39, + "content": "( W )", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 245, + 449, + 257 + ], + "score": 1.0, + "content": "on the 6 source/target pairs of the Office dataset (Saenko et al., 2010).", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "table_body", + "bbox": [ + 110, + 260, + 503, + 367 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 110, + 260, + 503, + 367 + ], + "spans": [ + { + "bbox": [ + 110, + 260, + 503, + 367 + ], + "score": 0.979, + "html": "
MethodA→D A→W W→A W→D D→AD→W Avg.
FRODA-SVM FRODA-NN88.0 83.978.7 69.576.5 75.098.0 97.773.7 69.094.6 83.984.9 79.8
D-FRODA-SVM D-FRODA-NN87.478.1 70.177.1 75.198.5 96.873.6 69.294.4 84.584.9 79.9
D-FRODA-W83.9 71.165.368.183.067.379.172.3
D-FRODA-U-SVM81.983.575.596.270.694.283.7
D-FRODA-U-NN78.172.169.1
93.667.075.775.9
D-FRODA-U-W73.465.962.888.061.765.169.5
", + "type": "table", + "image_path": "7371dcf571c6a2370810a973499090f8b06cbfba572289f2e1864fba2a9e6a42.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 110, + 260, + 503, + 295.6666666666667 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 110, + 295.6666666666667, + 503, + 331.33333333333337 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 110, + 331.33333333333337, + 503, + 367.00000000000006 + ], + "spans": [], + "index": 9 + } + ] + } + ], + "index": 6.75 + }, + { + "type": "text", + "bbox": [ + 106, + 393, + 505, + 427 + ], + "lines": [ + { + "bbox": [ + 105, + 393, + 506, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 506, + 406 + ], + "score": 1.0, + "content": "Fig. 1, we plot the average accuracy over all 12 pairs of the BCIS dataset as a function of the value of", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 405, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 505, + 417 + ], + "score": 1.0, + "content": "ε. Note that, once a sufficiently large threshold is reached, the results are quite stable. This indicates", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 416, + 393, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 416, + 393, + 428 + ], + "score": 1.0, + "content": "that our algorithm is robust to the specific value of this hyperparameter.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 106, + 446, + 505, + 513 + ], + "lines": [ + { + "bbox": [ + 105, + 445, + 506, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 506, + 460 + ], + "score": 1.0, + "content": "Results on the Office dataset. We further evaluate our approach on the slightly less challeng-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 458, + 506, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 458, + 506, + 470 + ], + "score": 1.0, + "content": "ing, although standard Office benchmark (Saenko et al., 2010). This dataset contains three different", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 469, + 505, + 481 + ], + "spans": [ + { + "bbox": [ + 106, + 469, + 505, + 481 + ], + "score": 1.0, + "content": "domains, namely Amazon (A), DSLR (D) and Webcam (W), sharing 31 object categories, but dif-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 480, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 106, + 480, + 505, + 493 + ], + "score": 1.0, + "content": "fering in data acquisition process. As in (Busto & Gall, 2017), we take all the samples from the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 490, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 106, + 490, + 428, + 503 + ], + "score": 1.0, + "content": "first 10 classes to represent the known ones, and all the samples with class labels", + "type": "text" + }, + { + "bbox": [ + 428, + 491, + 487, + 502 + ], + "score": 0.8, + "content": "1 1 , 1 2 , \\cdots , 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 490, + 505, + 503 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 107, + 501, + 361, + 515 + ], + "spans": [ + { + "bbox": [ + 107, + 502, + 166, + 514 + ], + "score": 0.84, + "content": "2 1 , 2 2 , \\cdots , 3 1", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 501, + 361, + 515 + ], + "score": 1.0, + "content": "as unknown source and target data, respectively.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 107, + 518, + 505, + 596 + ], + "lines": [ + { + "bbox": [ + 105, + 518, + 505, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 505, + 531 + ], + "score": 1.0, + "content": "We report the results of our algorithms and of the baselines for all 6 domain pairs of this dataset in", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 529, + 505, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 505, + 542 + ], + "score": 1.0, + "content": "Table 3. As before, note that we outperform the baselines in this open-set scenario. This includes the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 541, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 106, + 541, + 505, + 552 + ], + "score": 1.0, + "content": "end-to-end AlexNet-based approach of Saito et al. (2018) for open-set domain adaptation, as well", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 552, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 106, + 552, + 505, + 563 + ], + "score": 1.0, + "content": "as the state-of-the-art UDA methods of Tzeng et al. (2017) and Bousmalis et al. (2016), the latter", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 562, + 505, + 575 + ], + "spans": [ + { + "bbox": [ + 105, + 562, + 505, + 575 + ], + "score": 1.0, + "content": "of which is closest in spirit to our approach. In Table 4, we compare the different variants of our", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 573, + 505, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 505, + 586 + ], + "score": 1.0, + "content": "approach. The conclusions that one can draw from these results are similar to those for the BCIS", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 584, + 504, + 597 + ], + "spans": [ + { + "bbox": [ + 105, + 584, + 504, + 597 + ], + "score": 1.0, + "content": "dataset, thus showing that our method generalizes well across different domain adaptation datasets.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 107, + 601, + 505, + 635 + ], + "lines": [ + { + "bbox": [ + 105, + 600, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 366, + 613 + ], + "score": 1.0, + "content": "To evaluate the robustness of our method to the hyper-parameters", + "type": "text" + }, + { + "bbox": [ + 367, + 601, + 386, + 613 + ], + "score": 0.26, + "content": "\\alpha , \\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 600, + 405, + 613 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 406, + 601, + 417, + 612 + ], + "score": 0.88, + "content": "\\lambda _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 417, + 600, + 505, + 613 + ], + "score": 1.0, + "content": ", in Fig. 2, we plot the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 613, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 276, + 624 + ], + "score": 1.0, + "content": "average accuracy of D-FRODA-NN (with", + "type": "text" + }, + { + "bbox": [ + 276, + 613, + 302, + 623 + ], + "score": 0.88, + "content": "k = 3", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 613, + 505, + 624 + ], + "score": 1.0, + "content": ") over all 6 pairs of the Office dataset as a function", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 622, + 475, + 636 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 167, + 636 + ], + "score": 1.0, + "content": "of the value of", + "type": "text" + }, + { + "bbox": [ + 167, + 625, + 174, + 633 + ], + "score": 0.59, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 622, + 177, + 636 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 178, + 624, + 185, + 635 + ], + "score": 0.71, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 622, + 205, + 636 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 206, + 624, + 217, + 634 + ], + "score": 0.87, + "content": "\\lambda _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 622, + 475, + 636 + ], + "score": 1.0, + "content": ". Note that our results are stable for large ranges of these values.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 107, + 654, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "Runtimes. As mentioned in Section 3, one iteration of our approach takes on average 0.05 second,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 666, + 506, + 677 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 506, + 677 + ], + "score": 1.0, + "content": "and our algorithm typically takes around 50 iterations to converge. This yields a total runtime of", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 677, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 690 + ], + "score": 1.0, + "content": "roughly 2.5 seconds. By contrast, the publicly available implementation of the method of Busto &", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 686, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 686, + 506, + 702 + ], + "score": 1.0, + "content": "Gall (2017) takes on average 8 seconds per iteration and typically converges in 4 iterations, leading", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "to a total runtime of roughly 32 seconds. Note that these runtimes were measured on the same", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "computer and that both methods rely on the same input features. Therefore, our approach is not only", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 720, + 489, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 489, + 733 + ], + "score": 1.0, + "content": "significantly more accurate than (Busto & Gall, 2017), but also faster by an order of magnitude.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 32 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "spans": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "score": 1.0, + "content": "8", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 108, + 105, + 505, + 218 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 81, + 504, + 100 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 78, + 506, + 93 + ], + "spans": [ + { + "bbox": [ + 105, + 78, + 506, + 93 + ], + "score": 1.0, + "content": "Table 3: Recognition accuracies on the 6 source/target pairs of the Office dataset (Saenko et al., 2010) using a", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 91, + 321, + 100 + ], + "spans": [ + { + "bbox": [ + 106, + 91, + 321, + 100 + ], + "score": 1.0, + "content": "linear SVM classifier. A: Amazon, W: Webcam, D: DSLR.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table_body", + "bbox": [ + 108, + 105, + 505, + 218 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 105, + 505, + 218 + ], + "spans": [ + { + "bbox": [ + 108, + 105, + 505, + 218 + ], + "score": 0.98, + "html": "
MethodA→D A→WW→AW→DD→AD →WAvg.
LSVM72.657.549.298.845.188.568.6
DAN (Long et al.,2016a)77.672.560.898.35788.475.8
RTN (Long et al., 2016b)76.67362.498.857.28976.2
BP(Ganin & Lempitsky,2014)78.375.96498.757.689.877.4
ADDA (Tzeng et al.,2017)52.558.354.189.145.379.163.1
DSN (Bousmalis et al.,2016)58.357.255.179.358.170.263.0
ATI (Busto & Gall,2017)79.878.476.798.871.394.483.2
AODA (Saito et al.,2018)76.674.981.296.962.394.681.1
FRODA88.078.776.598.073.794.684.9
D-FRODA87.478.177.198.573.694.484.9
", + "type": "table", + "image_path": "6af5dcaeacd2773b50d2e1e0ede51fb7a086b5fe46272a7400190bc2ae5d4a5d.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 108, + 105, + 505, + 142.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 108, + 142.66666666666666, + 505, + 180.33333333333331 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 108, + 180.33333333333331, + 505, + 217.99999999999997 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 1.75 + }, + { + "type": "table", + "bbox": [ + 110, + 260, + 503, + 367 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 235, + 505, + 256 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 235, + 505, + 246 + ], + "spans": [ + { + "bbox": [ + 106, + 235, + 505, + 246 + ], + "score": 1.0, + "content": "Table 4: Recognition accuracies of variants of our approaches using a linear SVM, nearest neighbor (NN) and", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 245, + 449, + 257 + ], + "spans": [ + { + "bbox": [ + 106, + 245, + 178, + 257 + ], + "score": 1.0, + "content": "our linear classifier", + "type": "text" + }, + { + "bbox": [ + 179, + 245, + 196, + 255 + ], + "score": 0.39, + "content": "( W )", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 245, + 449, + 257 + ], + "score": 1.0, + "content": "on the 6 source/target pairs of the Office dataset (Saenko et al., 2010).", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "table_body", + "bbox": [ + 110, + 260, + 503, + 367 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 110, + 260, + 503, + 367 + ], + "spans": [ + { + "bbox": [ + 110, + 260, + 503, + 367 + ], + "score": 0.979, + "html": "
MethodA→D A→W W→A W→D D→AD→W Avg.
FRODA-SVM FRODA-NN88.0 83.978.7 69.576.5 75.098.0 97.773.7 69.094.6 83.984.9 79.8
D-FRODA-SVM D-FRODA-NN87.478.1 70.177.1 75.198.5 96.873.6 69.294.4 84.584.9 79.9
D-FRODA-W83.9 71.165.368.183.067.379.172.3
D-FRODA-U-SVM81.983.575.596.270.694.283.7
D-FRODA-U-NN78.172.169.1
93.667.075.775.9
D-FRODA-U-W73.465.962.888.061.765.169.5
", + "type": "table", + "image_path": "7371dcf571c6a2370810a973499090f8b06cbfba572289f2e1864fba2a9e6a42.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 110, + 260, + 503, + 295.6666666666667 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 110, + 295.6666666666667, + 503, + 331.33333333333337 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 110, + 331.33333333333337, + 503, + 367.00000000000006 + ], + "spans": [], + "index": 9 + } + ] + } + ], + "index": 6.75 + }, + { + "type": "text", + "bbox": [ + 106, + 393, + 505, + 427 + ], + "lines": [ + { + "bbox": [ + 105, + 393, + 506, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 506, + 406 + ], + "score": 1.0, + "content": "Fig. 1, we plot the average accuracy over all 12 pairs of the BCIS dataset as a function of the value of", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 405, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 505, + 417 + ], + "score": 1.0, + "content": "ε. Note that, once a sufficiently large threshold is reached, the results are quite stable. This indicates", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 416, + 393, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 416, + 393, + 428 + ], + "score": 1.0, + "content": "that our algorithm is robust to the specific value of this hyperparameter.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11, + "bbox_fs": [ + 105, + 393, + 506, + 428 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 446, + 505, + 513 + ], + "lines": [ + { + "bbox": [ + 105, + 445, + 506, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 506, + 460 + ], + "score": 1.0, + "content": "Results on the Office dataset. We further evaluate our approach on the slightly less challeng-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 458, + 506, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 458, + 506, + 470 + ], + "score": 1.0, + "content": "ing, although standard Office benchmark (Saenko et al., 2010). This dataset contains three different", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 469, + 505, + 481 + ], + "spans": [ + { + "bbox": [ + 106, + 469, + 505, + 481 + ], + "score": 1.0, + "content": "domains, namely Amazon (A), DSLR (D) and Webcam (W), sharing 31 object categories, but dif-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 480, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 106, + 480, + 505, + 493 + ], + "score": 1.0, + "content": "fering in data acquisition process. As in (Busto & Gall, 2017), we take all the samples from the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 490, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 106, + 490, + 428, + 503 + ], + "score": 1.0, + "content": "first 10 classes to represent the known ones, and all the samples with class labels", + "type": "text" + }, + { + "bbox": [ + 428, + 491, + 487, + 502 + ], + "score": 0.8, + "content": "1 1 , 1 2 , \\cdots , 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 490, + 505, + 503 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 107, + 501, + 361, + 515 + ], + "spans": [ + { + "bbox": [ + 107, + 502, + 166, + 514 + ], + "score": 0.84, + "content": "2 1 , 2 2 , \\cdots , 3 1", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 501, + 361, + 515 + ], + "score": 1.0, + "content": "as unknown source and target data, respectively.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 445, + 506, + 515 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 518, + 505, + 596 + ], + "lines": [ + { + "bbox": [ + 105, + 518, + 505, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 505, + 531 + ], + "score": 1.0, + "content": "We report the results of our algorithms and of the baselines for all 6 domain pairs of this dataset in", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 529, + 505, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 505, + 542 + ], + "score": 1.0, + "content": "Table 3. As before, note that we outperform the baselines in this open-set scenario. This includes the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 541, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 106, + 541, + 505, + 552 + ], + "score": 1.0, + "content": "end-to-end AlexNet-based approach of Saito et al. (2018) for open-set domain adaptation, as well", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 552, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 106, + 552, + 505, + 563 + ], + "score": 1.0, + "content": "as the state-of-the-art UDA methods of Tzeng et al. (2017) and Bousmalis et al. (2016), the latter", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 562, + 505, + 575 + ], + "spans": [ + { + "bbox": [ + 105, + 562, + 505, + 575 + ], + "score": 1.0, + "content": "of which is closest in spirit to our approach. In Table 4, we compare the different variants of our", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 573, + 505, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 505, + 586 + ], + "score": 1.0, + "content": "approach. The conclusions that one can draw from these results are similar to those for the BCIS", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 584, + 504, + 597 + ], + "spans": [ + { + "bbox": [ + 105, + 584, + 504, + 597 + ], + "score": 1.0, + "content": "dataset, thus showing that our method generalizes well across different domain adaptation datasets.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 22, + "bbox_fs": [ + 105, + 518, + 505, + 597 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 601, + 505, + 635 + ], + "lines": [ + { + "bbox": [ + 105, + 600, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 366, + 613 + ], + "score": 1.0, + "content": "To evaluate the robustness of our method to the hyper-parameters", + "type": "text" + }, + { + "bbox": [ + 367, + 601, + 386, + 613 + ], + "score": 0.26, + "content": "\\alpha , \\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 600, + 405, + 613 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 406, + 601, + 417, + 612 + ], + "score": 0.88, + "content": "\\lambda _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 417, + 600, + 505, + 613 + ], + "score": 1.0, + "content": ", in Fig. 2, we plot the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 613, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 276, + 624 + ], + "score": 1.0, + "content": "average accuracy of D-FRODA-NN (with", + "type": "text" + }, + { + "bbox": [ + 276, + 613, + 302, + 623 + ], + "score": 0.88, + "content": "k = 3", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 613, + 505, + 624 + ], + "score": 1.0, + "content": ") over all 6 pairs of the Office dataset as a function", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 622, + 475, + 636 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 167, + 636 + ], + "score": 1.0, + "content": "of the value of", + "type": "text" + }, + { + "bbox": [ + 167, + 625, + 174, + 633 + ], + "score": 0.59, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 622, + 177, + 636 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 178, + 624, + 185, + 635 + ], + "score": 0.71, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 622, + 205, + 636 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 206, + 624, + 217, + 634 + ], + "score": 0.87, + "content": "\\lambda _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 622, + 475, + 636 + ], + "score": 1.0, + "content": ". Note that our results are stable for large ranges of these values.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27, + "bbox_fs": [ + 105, + 600, + 505, + 636 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 654, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "Runtimes. As mentioned in Section 3, one iteration of our approach takes on average 0.05 second,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 666, + 506, + 677 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 506, + 677 + ], + "score": 1.0, + "content": "and our algorithm typically takes around 50 iterations to converge. This yields a total runtime of", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 677, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 690 + ], + "score": 1.0, + "content": "roughly 2.5 seconds. By contrast, the publicly available implementation of the method of Busto &", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 686, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 686, + 506, + 702 + ], + "score": 1.0, + "content": "Gall (2017) takes on average 8 seconds per iteration and typically converges in 4 iterations, leading", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "to a total runtime of roughly 32 seconds. Note that these runtimes were measured on the same", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "computer and that both methods rely on the same input features. Therefore, our approach is not only", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 720, + 489, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 489, + 733 + ], + "score": 1.0, + "content": "significantly more accurate than (Busto & Gall, 2017), but also faster by an order of magnitude.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 32, + "bbox_fs": [ + 105, + 655, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 107, + 78, + 502, + 182 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 78, + 502, + 182 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 78, + 502, + 182 + ], + "spans": [ + { + "bbox": [ + 107, + 78, + 502, + 182 + ], + "score": 0.967, + "type": "image", + "image_path": "7d2f29cf4001d94a3f3885b5178acbf97ec3cb8dc7e43ac7475f30850d87875b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 107, + 78, + 502, + 112.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 107, + 112.66666666666666, + 502, + 147.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 107, + 147.33333333333331, + 502, + 181.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 238, + 188, + 374, + 199 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 236, + 186, + 374, + 201 + ], + "spans": [ + { + "bbox": [ + 236, + 186, + 324, + 201 + ], + "score": 1.0, + "content": "Figure 2: Sensitivity to", + "type": "text" + }, + { + "bbox": [ + 324, + 190, + 332, + 198 + ], + "score": 0.65, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 186, + 334, + 201 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 335, + 189, + 342, + 199 + ], + "score": 0.66, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 186, + 360, + 201 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 360, + 189, + 371, + 198 + ], + "score": 0.84, + "content": "\\lambda _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 186, + 374, + 201 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 106, + 221, + 505, + 342 + ], + "lines": [ + { + "bbox": [ + 106, + 221, + 505, + 234 + ], + "spans": [ + { + "bbox": [ + 106, + 221, + 505, + 234 + ], + "score": 1.0, + "content": "Further discussion: The experimental setup used in (Saito et al., 2018; Busto & Gall, 2017) and", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 232, + 504, + 244 + ], + "spans": [ + { + "bbox": [ + 106, + 232, + 504, + 244 + ], + "score": 1.0, + "content": "our work for open-set DA relies on features extracted using a network pre-trained on ImageNet,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 243, + 505, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 505, + 256 + ], + "score": 1.0, + "content": "which in fact can be argued to already contain semantic information about some of the unknown", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 254, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 505, + 266 + ], + "score": 1.0, + "content": "classes. To evidence that our approach does not crucially depend on this information, and thus val-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 266, + 504, + 277 + ], + "spans": [ + { + "bbox": [ + 106, + 266, + 504, + 277 + ], + "score": 1.0, + "content": "idate our results, we observed that 4 of the unknown classes in the Office dataset, namely Tape dis-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 276, + 506, + 288 + ], + "spans": [ + { + "bbox": [ + 105, + 276, + 506, + 288 + ], + "score": 1.0, + "content": "penser, Stapler, Scissors, Punchers, do not appear in ImageNet. We therefore performed additional", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 288, + 506, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 506, + 299 + ], + "score": 1.0, + "content": "experiments with only these classes as unknown ones and the same 10 known classes as before. We", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 298, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 106, + 298, + 505, + 311 + ], + "score": 1.0, + "content": "compared the accuracy of our formulations against the SVM baseline and the open-set ATI method", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 309, + 504, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 309, + 448, + 321 + ], + "score": 1.0, + "content": "of Busto & Gall (2017). The gap with respect to both baselines remains large: SVM:", + "type": "text" + }, + { + "bbox": [ + 448, + 309, + 480, + 320 + ], + "score": 0.86, + "content": "7 6 . 0 1 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 309, + 504, + 321 + ], + "score": 1.0, + "content": ", ATI:", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 320, + 505, + 333 + ], + "spans": [ + { + "bbox": [ + 106, + 320, + 133, + 331 + ], + "score": 0.86, + "content": "7 7 . 4 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 320, + 188, + 333 + ], + "score": 1.0, + "content": ", D-FRODA:", + "type": "text" + }, + { + "bbox": [ + 188, + 320, + 215, + 331 + ], + "score": 0.87, + "content": "78 . 2 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 320, + 277, + 333 + ], + "score": 1.0, + "content": ", and FRODA:", + "type": "text" + }, + { + "bbox": [ + 278, + 320, + 304, + 331 + ], + "score": 0.87, + "content": "78 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 320, + 505, + 333 + ], + "score": 1.0, + "content": ". This confirms that our method applies to truly", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 331, + 213, + 342 + ], + "spans": [ + { + "bbox": [ + 106, + 331, + 213, + 342 + ], + "score": 1.0, + "content": "never-seen-before classes.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 107, + 348, + 505, + 392 + ], + "lines": [ + { + "bbox": [ + 106, + 348, + 505, + 359 + ], + "spans": [ + { + "bbox": [ + 106, + 348, + 505, + 359 + ], + "score": 1.0, + "content": "Note that among the 15 unknown classes in the setup for BCIS, 8 of them are not shared with", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 359, + 505, + 371 + ], + "spans": [ + { + "bbox": [ + 106, + 359, + 505, + 371 + ], + "score": 1.0, + "content": "ImageNet, namely Windmill, Steering wheel, Can-soda, Sneaker, Skyscraper, Ladder, Motorcycle,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 370, + 505, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 505, + 382 + ], + "score": 1.0, + "content": "and Palm tree. This further confirms that our method handles the cases where no categorical or", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 381, + 341, + 393 + ], + "spans": [ + { + "bbox": [ + 106, + 381, + 341, + 393 + ], + "score": 1.0, + "content": "semantic information is available in the extracted features.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16.5 + }, + { + "type": "title", + "bbox": [ + 108, + 409, + 195, + 422 + ], + "lines": [ + { + "bbox": [ + 104, + 408, + 197, + 425 + ], + "spans": [ + { + "bbox": [ + 104, + 408, + 197, + 425 + ], + "score": 1.0, + "content": "5 CONCLUSION", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 435, + 505, + 545 + ], + "lines": [ + { + "bbox": [ + 106, + 434, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 106, + 434, + 506, + 447 + ], + "score": 1.0, + "content": "We have introduced a novel approach to open-set domain adaptation, based on the intuition that", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 446, + 505, + 458 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 505, + 458 + ], + "score": 1.0, + "content": "source and target samples coming from the same, known classes can be represented by a shared sub-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 456, + 506, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 506, + 470 + ], + "score": 1.0, + "content": "space, while target samples from unknown classes should be modeled with a private subspace. Each", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 468, + 505, + 481 + ], + "spans": [ + { + "bbox": [ + 105, + 468, + 505, + 481 + ], + "score": 1.0, + "content": "step of the resulting algorithms can be solved efficiently. As demonstrated by our experiments, our", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 479, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 505, + 491 + ], + "score": 1.0, + "content": "method outperforms the state of the art in open-set domain adaptation and is one order of magnitude", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 491, + 504, + 501 + ], + "spans": [ + { + "bbox": [ + 106, + 491, + 504, + 501 + ], + "score": 1.0, + "content": "faster than the technique of Busto & Gall (2017). We believe that this clearly evidences the benefits", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 501, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 106, + 501, + 505, + 513 + ], + "score": 1.0, + "content": "of learning factorized representations, which allows us to jointly discard the unknown target samples", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 511, + 505, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 505, + 524 + ], + "score": 1.0, + "content": "and learn a better shared representation. In the future, we will investigate ways to make better use", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 523, + 505, + 535 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 505, + 535 + ], + "score": 1.0, + "content": "of unknown source data, and to exploit more effective classifiers, such as SVM, directly within our", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 533, + 204, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 204, + 545 + ], + "score": 1.0, + "content": "D-FRODA formulation.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 24.5 + }, + { + "type": "title", + "bbox": [ + 107, + 563, + 175, + 575 + ], + "lines": [ + { + "bbox": [ + 106, + 564, + 176, + 576 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 176, + 576 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 104, + 582, + 506, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 581, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 505, + 596 + ], + "score": 1.0, + "content": "M. Baktashmotlagh, M. Harandi, B. Lovell, and M. Salzmann. Unsupervised domain adaptation by", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 592, + 442, + 606 + ], + "spans": [ + { + "bbox": [ + 115, + 592, + 442, + 606 + ], + "score": 1.0, + "content": "domain invariant projection. In Proc. Int. Conference on Computer Vision, 2013.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 613, + 506, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 506, + 627 + ], + "score": 1.0, + "content": "M. Baktashmotlagh, M. Harandi, B. Lovell, and M. Salzmann. Domain adaptation on statistical", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 625, + 475, + 639 + ], + "spans": [ + { + "bbox": [ + 115, + 625, + 475, + 639 + ], + "score": 1.0, + "content": "manifold. In Proc. IEEE Conference on Computer Vision and Pattern Recognition, 2014.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 644, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 505, + 661 + ], + "score": 1.0, + "content": "A. Bendale and T. Boult. Towards open world recognition. In Proc. IEEE Conference on Computer", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 116, + 657, + 271, + 669 + ], + "spans": [ + { + "bbox": [ + 116, + 657, + 271, + 669 + ], + "score": 1.0, + "content": "Vision and Pattern Recognition, 2015.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 677, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 691 + ], + "score": 1.0, + "content": "K. Bousmalis, G. Trigeorgis, N. Silberman, D. Krishnan, and D. Erhan. Domain separation net-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 116, + 689, + 420, + 702 + ], + "spans": [ + { + "bbox": [ + 116, + 689, + 420, + 702 + ], + "score": 1.0, + "content": "works. In Proc. Advances in Neural Information Processing Systems, 2016.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 104, + 708, + 506, + 725 + ], + "spans": [ + { + "bbox": [ + 104, + 708, + 506, + 725 + ], + "score": 1.0, + "content": "P. Busto and J. Gall. Open set domain adaptation. In Proc. Int. Conference on Computer Vision,", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 720, + 143, + 733 + ], + "spans": [ + { + "bbox": [ + 115, + 720, + 143, + 733 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 35.5 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 107, + 78, + 502, + 182 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 78, + 502, + 182 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 78, + 502, + 182 + ], + "spans": [ + { + "bbox": [ + 107, + 78, + 502, + 182 + ], + "score": 0.967, + "type": "image", + "image_path": "7d2f29cf4001d94a3f3885b5178acbf97ec3cb8dc7e43ac7475f30850d87875b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 107, + 78, + 502, + 112.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 107, + 112.66666666666666, + 502, + 147.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 107, + 147.33333333333331, + 502, + 181.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 238, + 188, + 374, + 199 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 236, + 186, + 374, + 201 + ], + "spans": [ + { + "bbox": [ + 236, + 186, + 324, + 201 + ], + "score": 1.0, + "content": "Figure 2: Sensitivity to", + "type": "text" + }, + { + "bbox": [ + 324, + 190, + 332, + 198 + ], + "score": 0.65, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 186, + 334, + 201 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 335, + 189, + 342, + 199 + ], + "score": 0.66, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 186, + 360, + 201 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 360, + 189, + 371, + 198 + ], + "score": 0.84, + "content": "\\lambda _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 186, + 374, + 201 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 106, + 221, + 505, + 342 + ], + "lines": [ + { + "bbox": [ + 106, + 221, + 505, + 234 + ], + "spans": [ + { + "bbox": [ + 106, + 221, + 505, + 234 + ], + "score": 1.0, + "content": "Further discussion: The experimental setup used in (Saito et al., 2018; Busto & Gall, 2017) and", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 232, + 504, + 244 + ], + "spans": [ + { + "bbox": [ + 106, + 232, + 504, + 244 + ], + "score": 1.0, + "content": "our work for open-set DA relies on features extracted using a network pre-trained on ImageNet,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 243, + 505, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 505, + 256 + ], + "score": 1.0, + "content": "which in fact can be argued to already contain semantic information about some of the unknown", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 254, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 505, + 266 + ], + "score": 1.0, + "content": "classes. To evidence that our approach does not crucially depend on this information, and thus val-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 266, + 504, + 277 + ], + "spans": [ + { + "bbox": [ + 106, + 266, + 504, + 277 + ], + "score": 1.0, + "content": "idate our results, we observed that 4 of the unknown classes in the Office dataset, namely Tape dis-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 276, + 506, + 288 + ], + "spans": [ + { + "bbox": [ + 105, + 276, + 506, + 288 + ], + "score": 1.0, + "content": "penser, Stapler, Scissors, Punchers, do not appear in ImageNet. We therefore performed additional", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 288, + 506, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 506, + 299 + ], + "score": 1.0, + "content": "experiments with only these classes as unknown ones and the same 10 known classes as before. We", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 298, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 106, + 298, + 505, + 311 + ], + "score": 1.0, + "content": "compared the accuracy of our formulations against the SVM baseline and the open-set ATI method", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 309, + 504, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 309, + 448, + 321 + ], + "score": 1.0, + "content": "of Busto & Gall (2017). The gap with respect to both baselines remains large: SVM:", + "type": "text" + }, + { + "bbox": [ + 448, + 309, + 480, + 320 + ], + "score": 0.86, + "content": "7 6 . 0 1 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 309, + 504, + 321 + ], + "score": 1.0, + "content": ", ATI:", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 320, + 505, + 333 + ], + "spans": [ + { + "bbox": [ + 106, + 320, + 133, + 331 + ], + "score": 0.86, + "content": "7 7 . 4 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 320, + 188, + 333 + ], + "score": 1.0, + "content": ", D-FRODA:", + "type": "text" + }, + { + "bbox": [ + 188, + 320, + 215, + 331 + ], + "score": 0.87, + "content": "78 . 2 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 320, + 277, + 333 + ], + "score": 1.0, + "content": ", and FRODA:", + "type": "text" + }, + { + "bbox": [ + 278, + 320, + 304, + 331 + ], + "score": 0.87, + "content": "78 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 320, + 505, + 333 + ], + "score": 1.0, + "content": ". This confirms that our method applies to truly", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 331, + 213, + 342 + ], + "spans": [ + { + "bbox": [ + 106, + 331, + 213, + 342 + ], + "score": 1.0, + "content": "never-seen-before classes.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 221, + 506, + 342 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 348, + 505, + 392 + ], + "lines": [ + { + "bbox": [ + 106, + 348, + 505, + 359 + ], + "spans": [ + { + "bbox": [ + 106, + 348, + 505, + 359 + ], + "score": 1.0, + "content": "Note that among the 15 unknown classes in the setup for BCIS, 8 of them are not shared with", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 359, + 505, + 371 + ], + "spans": [ + { + "bbox": [ + 106, + 359, + 505, + 371 + ], + "score": 1.0, + "content": "ImageNet, namely Windmill, Steering wheel, Can-soda, Sneaker, Skyscraper, Ladder, Motorcycle,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 370, + 505, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 505, + 382 + ], + "score": 1.0, + "content": "and Palm tree. This further confirms that our method handles the cases where no categorical or", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 381, + 341, + 393 + ], + "spans": [ + { + "bbox": [ + 106, + 381, + 341, + 393 + ], + "score": 1.0, + "content": "semantic information is available in the extracted features.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 348, + 505, + 393 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 409, + 195, + 422 + ], + "lines": [ + { + "bbox": [ + 104, + 408, + 197, + 425 + ], + "spans": [ + { + "bbox": [ + 104, + 408, + 197, + 425 + ], + "score": 1.0, + "content": "5 CONCLUSION", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 435, + 505, + 545 + ], + "lines": [ + { + "bbox": [ + 106, + 434, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 106, + 434, + 506, + 447 + ], + "score": 1.0, + "content": "We have introduced a novel approach to open-set domain adaptation, based on the intuition that", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 446, + 505, + 458 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 505, + 458 + ], + "score": 1.0, + "content": "source and target samples coming from the same, known classes can be represented by a shared sub-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 456, + 506, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 506, + 470 + ], + "score": 1.0, + "content": "space, while target samples from unknown classes should be modeled with a private subspace. Each", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 468, + 505, + 481 + ], + "spans": [ + { + "bbox": [ + 105, + 468, + 505, + 481 + ], + "score": 1.0, + "content": "step of the resulting algorithms can be solved efficiently. As demonstrated by our experiments, our", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 479, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 505, + 491 + ], + "score": 1.0, + "content": "method outperforms the state of the art in open-set domain adaptation and is one order of magnitude", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 491, + 504, + 501 + ], + "spans": [ + { + "bbox": [ + 106, + 491, + 504, + 501 + ], + "score": 1.0, + "content": "faster than the technique of Busto & Gall (2017). We believe that this clearly evidences the benefits", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 501, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 106, + 501, + 505, + 513 + ], + "score": 1.0, + "content": "of learning factorized representations, which allows us to jointly discard the unknown target samples", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 511, + 505, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 505, + 524 + ], + "score": 1.0, + "content": "and learn a better shared representation. In the future, we will investigate ways to make better use", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 523, + 505, + 535 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 505, + 535 + ], + "score": 1.0, + "content": "of unknown source data, and to exploit more effective classifiers, such as SVM, directly within our", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 533, + 204, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 204, + 545 + ], + "score": 1.0, + "content": "D-FRODA formulation.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 434, + 506, + 545 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 563, + 175, + 575 + ], + "lines": [ + { + "bbox": [ + 106, + 564, + 176, + 576 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 176, + 576 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "list", + "bbox": [ + 104, + 582, + 506, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 581, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 505, + 596 + ], + "score": 1.0, + "content": "M. Baktashmotlagh, M. Harandi, B. Lovell, and M. Salzmann. Unsupervised domain adaptation by", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 592, + 442, + 606 + ], + "spans": [ + { + "bbox": [ + 115, + 592, + 442, + 606 + ], + "score": 1.0, + "content": "domain invariant projection. In Proc. Int. Conference on Computer Vision, 2013.", + "type": "text" + } + ], + "index": 32, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 613, + 506, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 506, + 627 + ], + "score": 1.0, + "content": "M. Baktashmotlagh, M. Harandi, B. Lovell, and M. Salzmann. Domain adaptation on statistical", + "type": "text" + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 625, + 475, + 639 + ], + "spans": [ + { + "bbox": [ + 115, + 625, + 475, + 639 + ], + "score": 1.0, + "content": "manifold. In Proc. IEEE Conference on Computer Vision and Pattern Recognition, 2014.", + "type": "text" + } + ], + "index": 34, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 644, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 505, + 661 + ], + "score": 1.0, + "content": "A. Bendale and T. Boult. Towards open world recognition. In Proc. IEEE Conference on Computer", + "type": "text" + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 116, + 657, + 271, + 669 + ], + "spans": [ + { + "bbox": [ + 116, + 657, + 271, + 669 + ], + "score": 1.0, + "content": "Vision and Pattern Recognition, 2015.", + "type": "text" + } + ], + "index": 36, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 677, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 691 + ], + "score": 1.0, + "content": "K. Bousmalis, G. Trigeorgis, N. Silberman, D. Krishnan, and D. Erhan. Domain separation net-", + "type": "text" + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 116, + 689, + 420, + 702 + ], + "spans": [ + { + "bbox": [ + 116, + 689, + 420, + 702 + ], + "score": 1.0, + "content": "works. In Proc. Advances in Neural Information Processing Systems, 2016.", + "type": "text" + } + ], + "index": 38, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 708, + 506, + 725 + ], + "spans": [ + { + "bbox": [ + 104, + 708, + 506, + 725 + ], + "score": 1.0, + "content": "P. Busto and J. Gall. Open set domain adaptation. In Proc. Int. Conference on Computer Vision,", + "type": "text" + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 720, + 143, + 733 + ], + "spans": [ + { + "bbox": [ + 115, + 720, + 143, + 733 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 40, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "Balakrishnan S. Chopra S. and Gopalan R. Dlid: Deep learning for domain adaptation by interpo-", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 93, + 495, + 108 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 495, + 108 + ], + "score": 1.0, + "content": "lating between domains. In ICML workshop on Challenges in Representation Learning, 2013.", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 103, + 111, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 103, + 111, + 506, + 128 + ], + "score": 1.0, + "content": "J. Donahue, Y. Jia, O. Vinyals, J. Hoffman, N. Zhang, E. Tzeng, and T. Darrell. Decaf: A deep con-", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 124, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 115, + 124, + 505, + 138 + ], + "score": 1.0, + "content": "volutional activation feature for generic visual recognition. In Proc. Int. Conference on Machine", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 135, + 184, + 149 + ], + "spans": [ + { + "bbox": [ + 115, + 135, + 184, + 149 + ], + "score": 1.0, + "content": "Learning, 2013.", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 154, + 506, + 168 + ], + "spans": [ + { + "bbox": [ + 104, + 154, + 506, + 168 + ], + "score": 1.0, + "content": "B. Fernando, A. Habrard, M. Sebban, and T. Tuytelaars. Unsupervised visual domain adaptation", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 166, + 434, + 178 + ], + "spans": [ + { + "bbox": [ + 115, + 166, + 434, + 178 + ], + "score": 1.0, + "content": "using subspace alignment. In Proc. Int. Conference on Computer Vision, 2013.", + "type": "text", + "cross_page": true + } + ], + "index": 6, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 184, + 506, + 200 + ], + "spans": [ + { + "bbox": [ + 105, + 184, + 506, + 200 + ], + "score": 1.0, + "content": "Y. Ganin and V. Lempitsky. Unsupervised domain adaptation by backpropagation. arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 198, + 216, + 209 + ], + "spans": [ + { + "bbox": [ + 115, + 198, + 216, + 209 + ], + "score": 1.0, + "content": "arXiv:1409.7495, 2014.", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 216, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 105, + 216, + 505, + 230 + ], + "score": 1.0, + "content": "M. Ghifary, Bastiaan W. Kleijn, M. Zhang, D. Balduzzi, and W. Li. Deep reconstruction-", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 114, + 227, + 506, + 241 + ], + "spans": [ + { + "bbox": [ + 114, + 227, + 506, + 241 + ], + "score": 1.0, + "content": "classification networks for unsupervised domain adaptation. In Proc. European Conference on", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 238, + 214, + 252 + ], + "spans": [ + { + "bbox": [ + 115, + 238, + 214, + 252 + ], + "score": 1.0, + "content": "Computer Vision, 2016.", + "type": "text", + "cross_page": true + } + ], + "index": 11, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 258, + 505, + 271 + ], + "spans": [ + { + "bbox": [ + 105, + 258, + 505, + 271 + ], + "score": 1.0, + "content": "B. Gong, Y. Shi, F. Sha, and K. Grauman. Geodesic flow kernel for unsupervised domain adaptation.", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 270, + 432, + 282 + ], + "spans": [ + { + "bbox": [ + 115, + 270, + 432, + 282 + ], + "score": 1.0, + "content": "In Proc. IEEE Conference on Computer Vision and Pattern Recognition, 2012.", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 288, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 505, + 302 + ], + "score": 1.0, + "content": "B. Gong, K. Grauman, and F. Sha. Connecting the dots with landmarks: Discriminatively learn-", + "type": "text", + "cross_page": true + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 299, + 506, + 313 + ], + "spans": [ + { + "bbox": [ + 115, + 299, + 506, + 313 + ], + "score": 1.0, + "content": "ing domain-invariant features for unsupervised domain adaptation. In Proc. Int. Conference on", + "type": "text", + "cross_page": true + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 310, + 221, + 324 + ], + "spans": [ + { + "bbox": [ + 115, + 310, + 221, + 324 + ], + "score": 1.0, + "content": "Machine Learning, 2013.", + "type": "text", + "cross_page": true + } + ], + "index": 16, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 329, + 506, + 344 + ], + "spans": [ + { + "bbox": [ + 104, + 329, + 506, + 344 + ], + "score": 1.0, + "content": "R. Gopalan, R. Li, and R. Chellappa. Unsupervised adaptation across domain shifts by generating", + "type": "text", + "cross_page": true + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 341, + 505, + 355 + ], + "spans": [ + { + "bbox": [ + 115, + 341, + 505, + 355 + ], + "score": 1.0, + "content": "intermediate data representations. IEEE Transactions on Pattern Analysis and Machine Intelli-", + "type": "text", + "cross_page": true + } + ], + "index": 18 + }, + { + "bbox": [ + 114, + 353, + 170, + 365 + ], + "spans": [ + { + "bbox": [ + 114, + 353, + 170, + 365 + ], + "score": 1.0, + "content": "gence, 2014.", + "type": "text", + "cross_page": true + } + ], + "index": 19, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 370, + 505, + 387 + ], + "spans": [ + { + "bbox": [ + 104, + 370, + 505, + 387 + ], + "score": 1.0, + "content": "Y. Jia, M. Salzmann, and T. Darrell. Factorized latent spaces with structured sparsity. In Proc.", + "type": "text", + "cross_page": true + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 383, + 357, + 397 + ], + "spans": [ + { + "bbox": [ + 115, + 383, + 357, + 397 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems, 2010.", + "type": "text", + "cross_page": true + } + ], + "index": 21, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 401, + 506, + 416 + ], + "spans": [ + { + "bbox": [ + 104, + 401, + 506, + 416 + ], + "score": 1.0, + "content": "H. Lee, A. Battle, R. Raina, and A. Ng. Efficient sparse coding algorithms. In Proc. Advances in", + "type": "text", + "cross_page": true + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 414, + 304, + 428 + ], + "spans": [ + { + "bbox": [ + 115, + 414, + 304, + 428 + ], + "score": 1.0, + "content": "Neural Information Processing Systems, 2007.", + "type": "text", + "cross_page": true + } + ], + "index": 23, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 433, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 506, + 447 + ], + "score": 1.0, + "content": "M. Long, Y. Cao, J. Wang, and M. Jordan. Learning transferable features with deep adaptation", + "type": "text", + "cross_page": true + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 114, + 444, + 323, + 457 + ], + "spans": [ + { + "bbox": [ + 114, + 444, + 323, + 457 + ], + "score": 1.0, + "content": "networks. arXiv preprint arXiv:1502.02791, 2015.", + "type": "text", + "cross_page": true + } + ], + "index": 25, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 464, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 505, + 478 + ], + "score": 1.0, + "content": "M. Long, J. Wang, and M. Jordan. Deep transfer learning with joint adaptation networks. corr, vol.", + "type": "text", + "cross_page": true + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 475, + 284, + 489 + ], + "spans": [ + { + "bbox": [ + 115, + 475, + 284, + 489 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1605.06636, 2016a.", + "type": "text", + "cross_page": true + } + ], + "index": 27, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 493, + 506, + 508 + ], + "spans": [ + { + "bbox": [ + 104, + 493, + 506, + 508 + ], + "score": 1.0, + "content": "M. Long, H. Zhu, J. Wang, and M. Jordan. Unsupervised domain adaptation with residual transfer", + "type": "text", + "cross_page": true + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 506, + 438, + 520 + ], + "spans": [ + { + "bbox": [ + 115, + 506, + 438, + 520 + ], + "score": 1.0, + "content": "networks. In Proc. Advances in Neural Information Processing Systems, 2016b.", + "type": "text", + "cross_page": true + } + ], + "index": 29, + "is_list_end_line": true + }, + { + "bbox": [ + 103, + 523, + 507, + 540 + ], + "spans": [ + { + "bbox": [ + 103, + 523, + 507, + 540 + ], + "score": 1.0, + "content": "J. Mairal, F. Bach, J. Ponce, G. Sapiro, R. Jenatton, and G. Obozinski. Spams: A sparse modeling", + "type": "text", + "cross_page": true + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 537, + 439, + 549 + ], + "spans": [ + { + "bbox": [ + 115, + 537, + 439, + 549 + ], + "score": 1.0, + "content": "software, v2. 3. URL http://spams-devel. gforge. inria. fr/downloads.html, 2014.", + "type": "text", + "cross_page": true + } + ], + "index": 31, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 556, + 506, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 506, + 569 + ], + "score": 1.0, + "content": "S. Pan, I. Tsang, J. Kwok, and Q. Yang. Domain adaptation via transfer component analysis. IEEE", + "type": "text", + "cross_page": true + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 567, + 281, + 581 + ], + "spans": [ + { + "bbox": [ + 115, + 567, + 281, + 581 + ], + "score": 1.0, + "content": "Transactions on Neural Networks, 2011.", + "type": "text", + "cross_page": true + } + ], + "index": 33, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 586, + 506, + 600 + ], + "spans": [ + { + "bbox": [ + 104, + 586, + 506, + 600 + ], + "score": 1.0, + "content": "J Quiñonero C., M. Sugiyama, A. Schwaighofer, and N. Lawrence. Covariate shift by kernel mean", + "type": "text", + "cross_page": true + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 599, + 327, + 610 + ], + "spans": [ + { + "bbox": [ + 115, + 599, + 327, + 610 + ], + "score": 1.0, + "content": "matching. Dataset Shift in Machine Learning, 2009.", + "type": "text", + "cross_page": true + } + ], + "index": 35, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 617, + 506, + 631 + ], + "spans": [ + { + "bbox": [ + 104, + 617, + 506, + 631 + ], + "score": 1.0, + "content": "A. Rozantsev, M. Salzmann, and P. Fua. Beyond sharing weights for deep domain adaptation. IEEE", + "type": "text", + "cross_page": true + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 116, + 630, + 383, + 641 + ], + "spans": [ + { + "bbox": [ + 116, + 630, + 383, + 641 + ], + "score": 1.0, + "content": "Transactions on Pattern Analysis and Machine Intelligence, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 37, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 649, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 505, + 660 + ], + "score": 1.0, + "content": "K. Saenko, B. Kulis, M. Fritz, and T. Darrell. Adapting visual category models to new domains. In", + "type": "text", + "cross_page": true + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 660, + 339, + 672 + ], + "spans": [ + { + "bbox": [ + 115, + 660, + 339, + 672 + ], + "score": 1.0, + "content": "Proc. European Conference on Computer Vision, 2010.", + "type": "text", + "cross_page": true + } + ], + "index": 39, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 678, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 104, + 678, + 505, + 693 + ], + "score": 1.0, + "content": "K. Saito, S. Yamamoto, Y. Ushiku, and T. Harada. Open set domain adaptation by backpropagation.", + "type": "text", + "cross_page": true + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 690, + 340, + 704 + ], + "spans": [ + { + "bbox": [ + 115, + 690, + 340, + 704 + ], + "score": 1.0, + "content": "Proc. European Conference on Computer Vision, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 41, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "W. J Scheirer, L. Jain, and T. Boult. Probability models for open set recognition. IEEE Transactions", + "type": "text", + "cross_page": true + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 720, + 329, + 734 + ], + "spans": [ + { + "bbox": [ + 115, + 720, + 329, + 734 + ], + "score": 1.0, + "content": "on Pattern Analysis and Machine Intelligence, 2014.", + "type": "text", + "cross_page": true + } + ], + "index": 43, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 82, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 96 + ], + "score": 1.0, + "content": "B. Sun and K. Saenko. Deep coral: Correlation alignment for deep domain adaptation. In Proc.", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 117, + 95, + 315, + 106 + ], + "spans": [ + { + "bbox": [ + 117, + 95, + 315, + 106 + ], + "score": 1.0, + "content": "European Conference on Computer Vision, 2016.", + "type": "text", + "cross_page": true + } + ], + "index": 1, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 111, + 505, + 126 + ], + "spans": [ + { + "bbox": [ + 105, + 111, + 505, + 126 + ], + "score": 1.0, + "content": "B. Sun, J. Feng, and K. Saenko. Return of frustratingly easy domain adaptation. In AAAI Conference", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 117, + 124, + 244, + 136 + ], + "spans": [ + { + "bbox": [ + 117, + 124, + 244, + 136 + ], + "score": 1.0, + "content": "on Artificial Intelligence, 2016.", + "type": "text", + "cross_page": true + } + ], + "index": 3, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 141, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 141, + 505, + 156 + ], + "score": 1.0, + "content": "T. Tommasi and T. Tuytelaars. A testbed for cross-dataset analysis. In Proc. European Conference", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 116, + 154, + 226, + 166 + ], + "spans": [ + { + "bbox": [ + 116, + 154, + 226, + 166 + ], + "score": 1.0, + "content": "on Computer Vision, 2014.", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 170, + 506, + 187 + ], + "spans": [ + { + "bbox": [ + 104, + 170, + 506, + 187 + ], + "score": 1.0, + "content": "E. Tzeng, J. Hoffman, N. Zhang, K. Saenko, and T. Darrell. Deep domain confusion: Maximizing", + "type": "text", + "cross_page": true + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 116, + 183, + 367, + 196 + ], + "spans": [ + { + "bbox": [ + 116, + 183, + 367, + 196 + ], + "score": 1.0, + "content": "for domain invariance. arXiv preprint arXiv:1412.3474, 2014.", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 201, + 505, + 215 + ], + "spans": [ + { + "bbox": [ + 105, + 201, + 505, + 215 + ], + "score": 1.0, + "content": "E. Tzeng, J. Hoffman, T. Darrell, and K. Saenko. Simultaneous deep transfer across domains and", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 116, + 214, + 351, + 226 + ], + "spans": [ + { + "bbox": [ + 116, + 214, + 351, + 226 + ], + "score": 1.0, + "content": "tasks. In Proc. Int. Conference on Computer Vision, 2015.", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 232, + 505, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 232, + 505, + 245 + ], + "score": 1.0, + "content": "E. Tzeng, J. Hoffman, K. Saenko, and T. Darrell. Adversarial discriminative domain adaptation. In", + "type": "text", + "cross_page": true + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 243, + 422, + 257 + ], + "spans": [ + { + "bbox": [ + 115, + 243, + 422, + 257 + ], + "score": 1.0, + "content": "Proc. IEEE Conference on Computer Vision and Pattern Recognition, 2017.", + "type": "text", + "cross_page": true + } + ], + "index": 11, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 261, + 505, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 505, + 275 + ], + "score": 1.0, + "content": "H. Yan, Y. Ding, P. Li, Q. Wang, Y. Xu, and W. Zuo. Mind the class weight bias: Weighted maximum", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 273, + 503, + 286 + ], + "spans": [ + { + "bbox": [ + 115, + 273, + 503, + 286 + ], + "score": 1.0, + "content": "mean discrepancy for unsupervised domain adaptation. arXiv preprint arXiv:1705.00609, 2017.", + "type": "text", + "cross_page": true + } + ], + "index": 13 + } + ], + "index": 35.5, + "bbox_fs": [ + 104, + 581, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 104, + 68, + 507, + 735 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "Balakrishnan S. Chopra S. and Gopalan R. Dlid: Deep learning for domain adaptation by interpo-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 495, + 108 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 495, + 108 + ], + "score": 1.0, + "content": "lating between domains. In ICML workshop on Challenges in Representation Learning, 2013.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 103, + 111, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 103, + 111, + 506, + 128 + ], + "score": 1.0, + "content": "J. Donahue, Y. Jia, O. Vinyals, J. Hoffman, N. Zhang, E. Tzeng, and T. Darrell. Decaf: A deep con-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 124, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 115, + 124, + 505, + 138 + ], + "score": 1.0, + "content": "volutional activation feature for generic visual recognition. In Proc. Int. Conference on Machine", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 135, + 184, + 149 + ], + "spans": [ + { + "bbox": [ + 115, + 135, + 184, + 149 + ], + "score": 1.0, + "content": "Learning, 2013.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 104, + 154, + 506, + 168 + ], + "spans": [ + { + "bbox": [ + 104, + 154, + 506, + 168 + ], + "score": 1.0, + "content": "B. Fernando, A. Habrard, M. Sebban, and T. Tuytelaars. Unsupervised visual domain adaptation", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 166, + 434, + 178 + ], + "spans": [ + { + "bbox": [ + 115, + 166, + 434, + 178 + ], + "score": 1.0, + "content": "using subspace alignment. In Proc. Int. Conference on Computer Vision, 2013.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 184, + 506, + 200 + ], + "spans": [ + { + "bbox": [ + 105, + 184, + 506, + 200 + ], + "score": 1.0, + "content": "Y. Ganin and V. Lempitsky. Unsupervised domain adaptation by backpropagation. arXiv preprint", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 198, + 216, + 209 + ], + "spans": [ + { + "bbox": [ + 115, + 198, + 216, + 209 + ], + "score": 1.0, + "content": "arXiv:1409.7495, 2014.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 216, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 105, + 216, + 505, + 230 + ], + "score": 1.0, + "content": "M. Ghifary, Bastiaan W. Kleijn, M. Zhang, D. Balduzzi, and W. Li. Deep reconstruction-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 114, + 227, + 506, + 241 + ], + "spans": [ + { + "bbox": [ + 114, + 227, + 506, + 241 + ], + "score": 1.0, + "content": "classification networks for unsupervised domain adaptation. In Proc. European Conference on", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 238, + 214, + 252 + ], + "spans": [ + { + "bbox": [ + 115, + 238, + 214, + 252 + ], + "score": 1.0, + "content": "Computer Vision, 2016.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 258, + 505, + 271 + ], + "spans": [ + { + "bbox": [ + 105, + 258, + 505, + 271 + ], + "score": 1.0, + "content": "B. Gong, Y. Shi, F. Sha, and K. Grauman. Geodesic flow kernel for unsupervised domain adaptation.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 270, + 432, + 282 + ], + "spans": [ + { + "bbox": [ + 115, + 270, + 432, + 282 + ], + "score": 1.0, + "content": "In Proc. IEEE Conference on Computer Vision and Pattern Recognition, 2012.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 288, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 505, + 302 + ], + "score": 1.0, + "content": "B. Gong, K. Grauman, and F. Sha. Connecting the dots with landmarks: Discriminatively learn-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 299, + 506, + 313 + ], + "spans": [ + { + "bbox": [ + 115, + 299, + 506, + 313 + ], + "score": 1.0, + "content": "ing domain-invariant features for unsupervised domain adaptation. In Proc. Int. Conference on", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 310, + 221, + 324 + ], + "spans": [ + { + "bbox": [ + 115, + 310, + 221, + 324 + ], + "score": 1.0, + "content": "Machine Learning, 2013.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 329, + 506, + 344 + ], + "spans": [ + { + "bbox": [ + 104, + 329, + 506, + 344 + ], + "score": 1.0, + "content": "R. Gopalan, R. Li, and R. Chellappa. Unsupervised adaptation across domain shifts by generating", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 341, + 505, + 355 + ], + "spans": [ + { + "bbox": [ + 115, + 341, + 505, + 355 + ], + "score": 1.0, + "content": "intermediate data representations. IEEE Transactions on Pattern Analysis and Machine Intelli-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 114, + 353, + 170, + 365 + ], + "spans": [ + { + "bbox": [ + 114, + 353, + 170, + 365 + ], + "score": 1.0, + "content": "gence, 2014.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 104, + 370, + 505, + 387 + ], + "spans": [ + { + "bbox": [ + 104, + 370, + 505, + 387 + ], + "score": 1.0, + "content": "Y. Jia, M. Salzmann, and T. Darrell. Factorized latent spaces with structured sparsity. In Proc.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 383, + 357, + 397 + ], + "spans": [ + { + "bbox": [ + 115, + 383, + 357, + 397 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems, 2010.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 104, + 401, + 506, + 416 + ], + "spans": [ + { + "bbox": [ + 104, + 401, + 506, + 416 + ], + "score": 1.0, + "content": "H. Lee, A. Battle, R. Raina, and A. Ng. Efficient sparse coding algorithms. In Proc. Advances in", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 414, + 304, + 428 + ], + "spans": [ + { + "bbox": [ + 115, + 414, + 304, + 428 + ], + "score": 1.0, + "content": "Neural Information Processing Systems, 2007.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 433, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 506, + 447 + ], + "score": 1.0, + "content": "M. Long, Y. Cao, J. Wang, and M. Jordan. Learning transferable features with deep adaptation", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 114, + 444, + 323, + 457 + ], + "spans": [ + { + "bbox": [ + 114, + 444, + 323, + 457 + ], + "score": 1.0, + "content": "networks. arXiv preprint arXiv:1502.02791, 2015.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 464, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 505, + 478 + ], + "score": 1.0, + "content": "M. Long, J. Wang, and M. Jordan. Deep transfer learning with joint adaptation networks. corr, vol.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 475, + 284, + 489 + ], + "spans": [ + { + "bbox": [ + 115, + 475, + 284, + 489 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1605.06636, 2016a.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 493, + 506, + 508 + ], + "spans": [ + { + "bbox": [ + 104, + 493, + 506, + 508 + ], + "score": 1.0, + "content": "M. Long, H. Zhu, J. Wang, and M. Jordan. Unsupervised domain adaptation with residual transfer", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 506, + 438, + 520 + ], + "spans": [ + { + "bbox": [ + 115, + 506, + 438, + 520 + ], + "score": 1.0, + "content": "networks. In Proc. Advances in Neural Information Processing Systems, 2016b.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 103, + 523, + 507, + 540 + ], + "spans": [ + { + "bbox": [ + 103, + 523, + 507, + 540 + ], + "score": 1.0, + "content": "J. Mairal, F. Bach, J. Ponce, G. Sapiro, R. Jenatton, and G. Obozinski. Spams: A sparse modeling", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 537, + 439, + 549 + ], + "spans": [ + { + "bbox": [ + 115, + 537, + 439, + 549 + ], + "score": 1.0, + "content": "software, v2. 3. URL http://spams-devel. gforge. inria. fr/downloads.html, 2014.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 556, + 506, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 506, + 569 + ], + "score": 1.0, + "content": "S. Pan, I. Tsang, J. Kwok, and Q. Yang. Domain adaptation via transfer component analysis. IEEE", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 567, + 281, + 581 + ], + "spans": [ + { + "bbox": [ + 115, + 567, + 281, + 581 + ], + "score": 1.0, + "content": "Transactions on Neural Networks, 2011.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 104, + 586, + 506, + 600 + ], + "spans": [ + { + "bbox": [ + 104, + 586, + 506, + 600 + ], + "score": 1.0, + "content": "J Quiñonero C., M. Sugiyama, A. Schwaighofer, and N. Lawrence. Covariate shift by kernel mean", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 599, + 327, + 610 + ], + "spans": [ + { + "bbox": [ + 115, + 599, + 327, + 610 + ], + "score": 1.0, + "content": "matching. Dataset Shift in Machine Learning, 2009.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 617, + 506, + 631 + ], + "spans": [ + { + "bbox": [ + 104, + 617, + 506, + 631 + ], + "score": 1.0, + "content": "A. Rozantsev, M. Salzmann, and P. Fua. Beyond sharing weights for deep domain adaptation. IEEE", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 116, + 630, + 383, + 641 + ], + "spans": [ + { + "bbox": [ + 116, + 630, + 383, + 641 + ], + "score": 1.0, + "content": "Transactions on Pattern Analysis and Machine Intelligence, 2018.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 649, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 505, + 660 + ], + "score": 1.0, + "content": "K. Saenko, B. Kulis, M. Fritz, and T. Darrell. Adapting visual category models to new domains. In", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 660, + 339, + 672 + ], + "spans": [ + { + "bbox": [ + 115, + 660, + 339, + 672 + ], + "score": 1.0, + "content": "Proc. European Conference on Computer Vision, 2010.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 678, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 104, + 678, + 505, + 693 + ], + "score": 1.0, + "content": "K. Saito, S. Yamamoto, Y. Ushiku, and T. Harada. Open set domain adaptation by backpropagation.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 690, + 340, + 704 + ], + "spans": [ + { + "bbox": [ + 115, + 690, + 340, + 704 + ], + "score": 1.0, + "content": "Proc. European Conference on Computer Vision, 2018.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "W. J Scheirer, L. Jain, and T. Boult. Probability models for open set recognition. IEEE Transactions", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 720, + 329, + 734 + ], + "spans": [ + { + "bbox": [ + 115, + 720, + 329, + 734 + ], + "score": 1.0, + "content": "on Pattern Analysis and Machine Intelligence, 2014.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 21.5 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "10", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 104, + 68, + 507, + 735 + ], + "lines": [], + "index": 21.5, + "bbox_fs": [ + 103, + 81, + 507, + 734 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 81, + 507, + 286 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 96 + ], + "score": 1.0, + "content": "B. Sun and K. Saenko. Deep coral: Correlation alignment for deep domain adaptation. In Proc.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 117, + 95, + 315, + 106 + ], + "spans": [ + { + "bbox": [ + 117, + 95, + 315, + 106 + ], + "score": 1.0, + "content": "European Conference on Computer Vision, 2016.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 111, + 505, + 126 + ], + "spans": [ + { + "bbox": [ + 105, + 111, + 505, + 126 + ], + "score": 1.0, + "content": "B. Sun, J. Feng, and K. Saenko. Return of frustratingly easy domain adaptation. In AAAI Conference", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 117, + 124, + 244, + 136 + ], + "spans": [ + { + "bbox": [ + 117, + 124, + 244, + 136 + ], + "score": 1.0, + "content": "on Artificial Intelligence, 2016.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 141, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 141, + 505, + 156 + ], + "score": 1.0, + "content": "T. Tommasi and T. Tuytelaars. A testbed for cross-dataset analysis. In Proc. European Conference", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 154, + 226, + 166 + ], + "spans": [ + { + "bbox": [ + 116, + 154, + 226, + 166 + ], + "score": 1.0, + "content": "on Computer Vision, 2014.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 104, + 170, + 506, + 187 + ], + "spans": [ + { + "bbox": [ + 104, + 170, + 506, + 187 + ], + "score": 1.0, + "content": "E. Tzeng, J. Hoffman, N. Zhang, K. Saenko, and T. Darrell. Deep domain confusion: Maximizing", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 116, + 183, + 367, + 196 + ], + "spans": [ + { + "bbox": [ + 116, + 183, + 367, + 196 + ], + "score": 1.0, + "content": "for domain invariance. arXiv preprint arXiv:1412.3474, 2014.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 201, + 505, + 215 + ], + "spans": [ + { + "bbox": [ + 105, + 201, + 505, + 215 + ], + "score": 1.0, + "content": "E. Tzeng, J. Hoffman, T. Darrell, and K. Saenko. Simultaneous deep transfer across domains and", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 116, + 214, + 351, + 226 + ], + "spans": [ + { + "bbox": [ + 116, + 214, + 351, + 226 + ], + "score": 1.0, + "content": "tasks. In Proc. Int. Conference on Computer Vision, 2015.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 232, + 505, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 232, + 505, + 245 + ], + "score": 1.0, + "content": "E. Tzeng, J. Hoffman, K. Saenko, and T. Darrell. Adversarial discriminative domain adaptation. In", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 243, + 422, + 257 + ], + "spans": [ + { + "bbox": [ + 115, + 243, + 422, + 257 + ], + "score": 1.0, + "content": "Proc. IEEE Conference on Computer Vision and Pattern Recognition, 2017.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 261, + 505, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 505, + 275 + ], + "score": 1.0, + "content": "H. Yan, Y. Ding, P. Li, Q. Wang, Y. Xu, and W. Zuo. Mind the class weight bias: Weighted maximum", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 273, + 503, + 286 + ], + "spans": [ + { + "bbox": [ + 115, + 273, + 503, + 286 + ], + "score": 1.0, + "content": "mean discrepancy for unsupervised domain adaptation. arXiv preprint arXiv:1705.00609, 2017.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 6.5 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 105, + 81, + 507, + 286 + ], + "lines": [], + "index": 6.5, + "bbox_fs": [ + 104, + 82, + 506, + 286 + ], + "lines_deleted": true + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file