1<!DOCTYPE html> 2<html> 3 4 <head> 5 6 <meta charset="utf-8"> 7<meta name="viewport" content="width=device-width, initial-scale=1, shrink-to-fit=no"> 8<meta http-equiv="X-UA-Compatible" content="IE=edge"> 9 10<title> 11 12 Jingling Li 13 14 15 | publications 16 17</title> 18<meta name="description" content="Jingling's personal website 19"> 20 21<!-- Open Graph --> 22 23 24<!-- Bootstrap & MDB --> 25<link href="https://stackpath.bootstrapcdn.com/bootstrap/4.5.2/css/bootstrap.min.css" rel="stylesheet" integrity="sha512-MoRNloxbStBcD8z3M/2BmnT+rg4IsMxPkXaGh2zD6LGNNFE80W3onsAhRcMAMrSoyWL9xD7Ert0men7vR8LUZg==" crossorigin="anonymous"> 26<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/mdbootstrap/4.19.1/css/mdb.min.css" integrity="sha512-RO38pBRxYH3SoOprtPTD86JFOclM51/XTIdEPh5j8sj4tp8jmQIx26twG52UaLi//hQldfrh7e51WzP9wuP32Q==" crossorigin="anonymous" /> 27 28<!-- Fonts & Icons --> 29<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/5.14.0/css/all.min.css" integrity="sha512-1PKOgIY59xJ8Co8+NE6FZ+LOAZKjy+KY8iq0G4B3CyeY6wYHN3yt9PW0XpSriVlkMXe40PTKnXrLnZ9+fkDaog==" crossorigin="anonymous"> 30<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/academicons/1.9.0/css/academicons.min.css" integrity="sha512-W4yqoT1+8NLkinBLBZko+dFB2ZbHsYLDdr50VElllRcNt2Q4/GSs6u71UHKxB7S6JEMCp5Ve4xjh3eGQl/HRvg==" crossorigin="anonymous"> 31<link rel="stylesheet" type="text/css" href="https://fonts.googleapis.com/css?family=Roboto:300,400,500,700|Roboto+Slab:100,300,400,500,700|Material+Icons"> 32 33<!-- Code Syntax Highlighting --> 34<link rel="stylesheet" href="https://gitcdn.link/repo/jwarby/jekyll-pygments-themes/master/github.css" /> 35 36<!-- Styles --> 37 38<link rel="icon" href="data:image/svg+xml,<svg xmlns=%22http://www.w3.org/2000/svg%22 viewBox=%220 0 100 100%22><text y=%22.9em%22 font-size=%2290%22>ð</text></svg>"> 39 40<link rel="stylesheet" href="/assets/css/main.css"> 41<link rel="canonical" href="/publications/"> 42 43<!-- JQuery --> 44<!-- jQuery -->
45<script src="https://cdnjs.cloudflare.com/ajax/libs/jquery/3.5.1/jquery.min.js" integrity="sha512-bLT0Qm9VnAYZDflyKcBaQ2gg0hSYNQrJ8RilYldYQ1FxQYoCLtUjuuRuZo+fjqhx/qtq/1itJ0C2ejDxltZVFg==" crossorigin="anonymous"></script>
45 46 47 48<!-- Theming--> 49
50<script src="/assets/js/theme.js"></script>
vendor: 1 bytes, line 50
50
51<script src="/assets/js/dark_mode.js"></script>
51 52 53 54 55 56 57 58 59<!-- MathJax -->
60<script type="text/javascript"> 61 window.MathJax = { 62 tex: { 63 tags: 'ams' 64 } 65 }; 66</script>
vendor: 1 bytes, line 66
66
67<script defer type="text/javascript" id="MathJax-script" src="https://cdn.jsdelivr.net/npm/[email protected]/es5/tex-mml-chtml.js"></script>
vendor: 1 bytes, line 67
67
68<script defer src="https://polyfill.io/v3/polyfill.min.js?features=es6"></script>
68 69 70 71 </head> 72 73 <body class="fixed-top-nav "> 74 75 <!-- Header --> 76 77 <header> 78 79 <!-- Nav Bar --> 80 <nav id="navbar" class="navbar navbar-light navbar-expand-sm fixed-top"> 81 <div class="container"> 82 83 <a class="navbar-brand title font-weight-lighter" href="https://jinglingli.github.io/"> 84 <span class="font-weight-bold">Jingling</span> Li 85 </a> 86 87 <!-- Navbar Toggle --> 88 <button class="navbar-toggler collapsed ml-auto" type="button" data-toggle="collapse" data-target="#navbarNav" aria-controls="navbarNav" aria-expanded="false" aria-label="Toggle navigation"> 89 <span class="sr-only">Toggle navigation</span> 90 <span class="icon-bar top-bar"></span> 91 <span class="icon-bar middle-bar"></span> 92 <span class="icon-bar bottom-bar"></span> 93 </button> 94 <div class="collapse navbar-collapse text-right" id="navbarNav"> 95 <ul class="navbar-nav ml-auto flex-nowrap"> 96 <!-- About --> 97 <li class="nav-item "> 98 <a class="nav-link" href="/"> 99 about 100 101 </a> 102 </li> 103 104 <!-- Other pages --> 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 <li class="nav-item active"> 130 <a class="nav-link" href="/publications/"> 131 publications 132 133 <span class="sr-only">(current)</span> 134 135 </a> 136 </li> 137 138 139 140 141 142 143 <div class="toggle-container"> 144 <a id="light-toggle"> 145 <i class="fas fa-moon"></i> 146 <i class="fas fa-sun"></i> 147 </a> 148 </div> 149 150 </ul> 151 </div> 152 </div> 153 </nav> 154 155</header> 156 157 158 <!-- Content --> 159 160 <div class="container mt-5"> 161 <div class="post"> 162 163 <header class="post-header"> 164 <h1 class="post-title">publications</h1> 165 <p class="post-description">\* denotes equal contributions or alphabetical order.</p> 166 </header> 167 168 <article> 169 <div class="publications"> 170 171 172 <h2 class="year">2021</h2> 173 <ol class="bibliography"> 174<li> 175<div class="row"> 176 <div class="col-sm-2 abbr"> 177 178 179 <abbr class="badge">Neurips</abbr> 180 181 182 </div> 183 184 <div id="Li2021How" class="col-sm-8"> 185 186 <div class="title">How Does a Neural Networkâs Architecture Impact Its Robustness to Noisy Labels?</div> 187 <div class="author"> 188 189 190 191 192 193 194 195 196 197 <em>Li, Jingling</em>, 198 199 200 201 202 203 204 205 206 207 208 209 210 211 Zhang, Mozhi, 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 Xu, Keyulu, 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 Dickerson, John, 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 and Ba, Jimmy 257 258 259 260 261 262 </div> 263 264 <div class="periodical"> 265 266 <em>In Advances in Neural Information Processing Systems,</em> 267 268 269 2021 270 271 </div> 272 273 274 <div class="links"> 275 276 <a class="abstract btn btn-sm z-depth-0" role="button">Abs</a> 277 278 279 280 281 <a href="https://openreview.net/forum?id=Ir-WwGboFN-" class="btn btn-sm z-depth-0" role="button" target="_blank" rel="noopener noreferrer">HTML</a> 282 283 284 285 286 287 <a href="https://github.com/jinglingli/alignment_noisy_label" class="btn btn-sm z-depth-0" role="button" target="_blank" rel="noopener noreferrer">
287Code</a> 288 289 290 291 292 </div> 293 294 <!-- Hidden abstract block --> 295 296 <div class="abstract hidden"> 297 <p>Noisy labels are inevitable in large real-world datasets. In this work, we explore an area understudied by previous works â how the networkâs architecture impacts its robustness to noisy labels. We provide a formal framework connecting the robustness of a network to the alignments between its architecture and target/noise functions. Our framework measures a networkâs robustness via the predictive power in its representations â the test performance of a linear model trained on the learned representations using a small set of clean labels. We hypothesize that a network is more robust to noisy labels if its architecture is more aligned with the target function than the noise. To support our hypothesis, we provide both theoretical and empirical evidence across various neural network architectures and different domains. We also find that when the network is well-aligned with the target function, its predictive power in representations could improve upon state-of-the-art (SOTA) noisy-label-training methods in terms of test accuracy and even outperform sophisticated methods that use clean labels.</p> 298 </div> 299 300 301 <!-- Hidden bibtex block --> 302 303 </div> 304</div> 305</li> 306<li> 307<div class="row"> 308 <div class="col-sm-2 abbr"> 309 310 311 <abbr class="badge">Neurips</abbr> 312 313 314 </div> 315 316 <div id="Ding2021Vq" class="col-sm-8"> 317 318 <div class="title">VQ-GNN: A Universal Framework to Scale up Graph Neural Networks using Vector Quantization</div> 319 <div class="author"> 320 321 322 323 324 325 326 327 328 329 330 Ding, Mucong*, 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 Kong, Kezhi*, 346 347 348 349 350 351 352 353 354 355 356 357 358 359 <em>Li, Jingling</em>, 360 361 362 363 364 365 366 367 368 369 370 371 372 373 Zhu, Chen, 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 Dickerson, John P, 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 Huang, Furong, 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 and Goldstein, Tom 419 420 421 422 423 424 </div> 425 426 <div class="periodical"> 427 428 <em>In Advances in Neural Information Processing Systems</em> 429 430 431 2021 432 433 </div> 434 435 436 <div class="links"> 437 438 <a class="abstract btn btn-sm z-depth-0" role="button">Abs</a> 439 440 441 442 443 <a href="https://openreview.net/forum?id=EO-CQzgcIxd" class="btn btn-sm z-depth-0" role="button" target="_blank" rel="noopener noreferrer">HTML</a> 444 445 446 447 448 449 <a href="https://github.com/devnkong/VQ-GNN" class="btn btn-sm z-depth-0" role="button" target="_blank" rel="noopener noreferrer">Code</a> 450 451 452 453 454 </div> 455 456 <!-- Hidden abstract block --> 457 458 <div class="abstract hidden"> 459 <p>Most state-of-the-art Graph Neural Networks (GNNs) can be defined as a form of graph convolution which can be realized by message passing between direct neighbors or beyond. To scale such GNNs to large graphs, various neighbor-, layer-, or subgraph-sampling techniques are proposed to alleviate the "neighbor explosion" problem by considering only a small subset of messages passed to the nodes in a mini-batch. However, sampling-based methods are difficult to apply to GNNs that utilize many-hops-away or global context each layer, show unstable performance for different tasks and datasets, and do not speed up model inference. We propose a principled and fundamentally different approach, VQ-GNN, a universal framework to
459scale up any convolution-based GNNs using Vector Quantization (VQ) without compromising the performance. In contrast to sampling-based techniques, our approach can effectively preserve all the messages passed to a mini-batch of nodes by learning and updating a small number of quantized reference vectors of global node representations, using VQ within each GNN layer. Our framework avoids the "neighbor explosion" problem of GNNs using quantized representations combined with a low-rank version of the graph convolution matrix. We show that such a compact low-rank version of the gigantic convolution matrix is sufficient both theoretically and experimentally. In company with VQ, we design a novel approximated message passing algorithm and a nontrivial back-propagation rule for our framework. Experiments on various types of GNN backbones demonstrate the scalability and competitive performance of our framework on large-graph node classification and link prediction benchmarks.</p> 460 </div> 461 462 463 <!-- Hidden bibtex block --> 464 465 </div> 466</div> 467</li> 468<li> 469<div class="row"> 470 <div class="col-sm-2 abbr"> 471 472 473 <abbr class="badge">ICLR</abbr> 474 475 476 </div> 477 478 <div id="Xu2021How" class="col-sm-8"> 479 480 <div class="title">How Neural Networks Extrapolate: From Feedforward to Graph Neural Networks</div> 481 <div class="author"> 482 483 484 485 486 487 488 489 490 491 492 Xu, Keyulu, 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 Zhang, Mozhi, 508 509 510 511 512 513 514 515 516 517 518 519 520 521 <em>Li, Jingling</em>, 522 523 524 525 526 527 528 529 530 531 532 533 534 535 Du, Simon S., 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 Kawarabayashi, Ken-ichi, 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 and Jegelka, Stefanie 566 567 568 569 570 571 </div> 572 573 <div class="periodical"> 574 575 <em>In International Conference on Learning Representations (Oral)</em> 576 577 578 2021 579 580 </div> 581 582 583 <div class="links"> 584 585 <a class="abstract btn btn-sm z-depth-0" role="button">Abs</a> 586 587 588 589 590 <a href="https://openreview.net/forum?id=UH-cmocLJC" class="btn btn-sm z-depth-0" role="button" target="_blank" rel="noopener noreferrer">HTML</a> 591 592 593 594 595 596 <a href="https://github.com/jinglingli/nn-extrapolate" class="btn btn-sm z-depth-0" role="button" target="_blank" rel="noopener noreferrer">Code</a> 597 598 599 600 601 </div> 602 603 <!-- Hidden abstract block --> 604 605 <div class="abstract hidden"> 606 <p>We study how neural networks trained by gradient descent extrapolate, i.e., what they learn outside the support of the training distribution. Previous works report mixed empirical results when extrapolating with neural networks: while feedforward neural networks, a.k.a. multilayer perceptrons (MLPs), do not extrapolate well in certain simple tasks, Graph Neural Networks (GNNs) â structured networks with MLP modules â have shown some success in more complex tasks. Working towards a theoretical explanation, we identify conditions under which MLPs and GNNs extrapolate well. First, we quantify the observation that ReLU MLPs quickly converge to linear functions along any direction from the origin, which implies that ReLU MLPs do not extrapolate most nonlinear functions. But, they can provably learn a linear target function when the training distribution is sufficiently "diverse". Second, in connection to analyzing the successes and limitations of GNNs, these results suggest a hypothesis for which we provide theoretical and empirical evidence: the success of GNNs in extrapolating algorithmic tasks to new data (e.g., larger graphs or edge weights) relies on encoding task-s
606pecific non-linearities in the architecture or features. Our theoretical analysis builds on a connection of over-parameterized networks to the neural tangent kernel. Empirically, our theory holds across different training settings.</p> 607 </div> 608 609 610 <!-- Hidden bibtex block --> 611 612 </div> 613</div> 614</li> 615</ol> 616 617 <h2 class="year">2020</h2> 618 <ol class="bibliography"> 619<li> 620<div class="row"> 621 <div class="col-sm-2 abbr"> 622 623 624 <abbr class="badge">ICLR</abbr> 625 626 627 </div> 628 629 <div id="Xu2020What" class="col-sm-8"> 630 631 <div class="title">What Can Neural Networks Reason About?</div> 632 <div class="author"> 633 634 635 636 637 638 639 640 641 642 643 Xu, Keyulu, 644 645 646 647 648 649 650 651 652 653 654 655 656 657 <em>Li, Jingling</em>, 658 659 660 661 662 663 664 665 666 667 668 669 670 671 Zhang, Mozhi, 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 Du, Simon S., 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 Kawarabayashi, Ken-ichi, 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 and Jegelka, Stefanie 717 718 719 720 721 722 </div> 723 724 <div class="periodical"> 725 726 <em>In International Conference on Learning Representations (Spotlight)</em> 727 728 729 2020 730 731 </div> 732 733 734 <div class="links"> 735 736 <a class="abstract btn btn-sm z-depth-0" role="button">Abs</a> 737 738 739 740 741 <a href="https://openreview.net/forum?id=rJxbJeHFPS" class="btn btn-sm z-depth-0" role="button" target="_blank" rel="noopener noreferrer">HTML</a> 742 743 744 745 746 747 <a href="https://github.com/NNReasoning/What-Can-Neural-Networks-Reason-About" class="btn btn-sm z-depth-0" role="button" target="_blank" rel="noopener noreferrer">Code</a> 748 749 750 751 752 </div> 753 754 <!-- Hidden abstract block --> 755 756 <div class="abstract hidden"> 757 <p>Neural networks have succeeded in many reasoning tasks. Empirically, these tasks require specialized network structures, e.g., Graph Neural Networks (GNNs) perform well on many such tasks, but less structured networks fail. Theoretically, there is limited understanding of why and when a network structure generalizes better than others, although they have equal expressive power. In this paper, we develop a framework to characterize which reasoning tasks a network can learn well, by studying how well its computation structure aligns with the algorithmic structure of the relevant reasoning process. We formally define this algorithmic alignment and derive a sample complexity bound that decreases with better alignment. This framework offers an explanation for the empirical success of popular reasoning models, and suggests their limitations. As an example, we unify seemingly different reasoning tasks, such as intuitive physics, visual question answering, and shortest paths, via the lens of a powerful algorithmic paradigm, dynamic programming (DP). We show that GNNs align with DP and thus are expected to solve these tasks. On several reasoning tasks, our theory is supported by empirical results.</p> 758 </div> 759 760 761 <!-- Hidden bibtex block --> 762 763 </div> 764</div> 765</li> 766<li> 767<div class="row"> 768 <div class="col-sm-2 abbr"> 769 770 771 <abbr class="badge">AISTATS</abbr> 772 773 774 </div> 775 776 <div id="Li2020Understanding" class="col-sm-8"> 777 778 <div class="title">Understanding Generalization in Deep Learning via Tensor Methods</div> 779 <div class="author"> 780 781 782 783 784 785 786 787 788 789 <em>Li, Jingling</em>, 790 791 792 793 794 795 796 797 798 799 800 801 802 803 Sun, Yanchao, 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 Su, Jiahao, 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 Suzuki, Taiji, 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 and Huang, Furong 849 850 851 852 853 854 </div> 855 856 <div class="periodical"> 857 858 <em>In International Conference on Artificial Intelligence and Statistics</em> 859 860 861 2020 862 863 </div> 864 865 866 <div class="links"> 867 868 <a class="abstract btn btn-sm z-depth-0" role="button">Abs</a> 869 870 871 872 873 <a href="http://proceedings.mlr.press/v108/li20c/li20c.pdf" class="btn btn-sm z-depth-0" role="button" target="_blank" rel="noopener noreferrer">HTML</a> 874 875 876 877 878 879 880 881 882 </div> 883 884 <!-- Hidden abstract block --> 885 886 <div class="abstract hidden"> 887 <p>Deep neural networks generalize well on unseen data though the number of parameters often far exceeds the number of training examples. Recently proposed complexity measures have provided insights to understanding the generalizability in neural networks from perspectives of PAC-Bayes, robustness, overparametrization, compression and so on. In this work, we advance the understanding of the relations between the networkâs architecture and its generalizability from the compression perspective. Using tensor analysis, we propose a series of intuitive, data-dependent and easily-measurable properties that tightly characterize the compressibility and generalizability of neural networks; thus, in practice, our generalization bound outperforms the previous compression-based ones, especially for neural networks using tensors as their weight kernels (e.g. CNNs). Moreover, these intuitive measurements provide further insights into designing neural network architectures with properties favorable for better/guaranteed generalizability. Our experimental results demonstrate that through the proposed measurable properties, our generalization error bound matches the trend of the test error well. Our theoretical analysis further provides justifications for the empirical success and limitations of some widely-used tensor-based compression approaches. We also discover the improvements to the compressibility and robustness of current neural networks when incorporating tensor operations via our proposed layer-wise structure.</p> 888 </div> 889 890 891 <!-- Hidden bibtex block --> 892 893 </div> 894</div> 895</li> 896</ol> 897 898 <h2 class="year">2019</h2> 899 <ol class="bibliography"><li> 900<div class="row"> 901 <div class="col-sm-2 abbr"> 902 903 904 <abbr class="badge">TCS</abbr> 905 906 907 </div> 908 909 <div id="khuller2019select" class="col-sm-8"> 910 911 <div class="title">Select and permute: An improved online framework for scheduling to minimize weighted completion time</div> 912 <div class="author"> 913 914 915 916 917 918 919 920 921 922 923 Khuller, Samir*, 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 Li, Jingling*, 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 Sturmfels, Pascal*, 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 Sun, Kevin*, 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 and Venkat, Prayaag* 984 985 986 987 988 989 </div> 990 991 <div class="periodical"> 992 993 <em>Theoretical Computer Science</em> 994 995 996 2019 997 998 </div> 999 1000 1001 <div class="links"> 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 </div> 1014 1015 <!-- Hidden abstract block --> 1016 1017 1018 <!-- Hidden bibtex block --> 1019 1020 </div> 1021</div> 1022</li></ol> 1023 1024 1025</div> 1026 1027 </article> 1028 1029</div> 1030 1031 </div> 1032 1033 <!-- Footer --> 1034 1035 1036<footer class="fixed-bottom"> 1037 <div class="container mt-0"> 1038 © Copyright 2022 Jingling Li. 1039 Powered by <a href="http://jekyllrb.com/" target="_blank" rel="noopener noreferrer">Jekyll</a> with <a href="https://github.com/alshedivat/al-folio" target="_blank" rel="noopener noreferrer">al-folio</a> theme. Hosted by <a href="https://pages.github.com/" target="_blank" rel="noopener noreferrer">GitHub Pages</a>. 1040 1041 1042 1043 Last updated: November 22, 2022. 1044 1045 </div> 1046</footer> 1047 1048 1049 1050 </body> 1051 1052 <!-- Bootsrap & MDB scripts -->
1053<script src="https://cdnjs.cloudflare.com/ajax/libs/popper.js/2.4.4/umd/popper.min.js" integrity="sha512-eUQ9hGdLjBjY3F41CScH3UX+4JDSI9zXeroz7hJ+RteoCaY+GP/LDoM8AO+Pt+DRFw3nXqsjh9Zsts8hnYv8/A==" crossorigin="anonymous"></script>
vendor: 1 bytes, line 1053
1053
1054<script src="https://stackpath.bootstrapcdn.com/bootstrap/4.5.2/js/bootstrap.min.js" integrity="sha512-M5KW3ztuIICmVIhjSqXe01oV2bpe248gOxqmlcYrEzAvws7Pw3z6BK0iGbrwvdrUQUhi3eXgtxp5I8PDo9YfjQ==" crossorigin="anonymous"></script>
vendor: 1 bytes, line 1054
1054
1055<script src="https://cdnjs.cloudflare.com/ajax/libs/mdbootstrap/4.19.1/js/mdb.min.js" integrity="sha512-Mug9KHKmroQFMLm93zGrjhibM2z2Obg9l6qFG2qKjXEXkMp/VDkI4uju9m4QKPjWSwQ6O2qzZEnJDEeCw0Blcw==" crossorigin="anonymous"></script>
1055 1056 1057 1058<!-- Mansory & imagesLoaded -->
1059<script defer src="https://unpkg.com/masonry-layout@4/dist/masonry.pkgd.min.js"></script>
vendor: 1 bytes, line 1059
1059
1060<script defer src="https://unpkg.com/imagesloaded@4/imagesloaded.pkgd.min.js"></script>
vendor: 1 bytes, line 1060
1060
1061<script defer src="/assets/js/mansory.js" type="text/javascript"></script>
1061 1062 1063 1064 1065 1066 1067<!-- Medium Zoom JS -->
1068<script src="https://cdn.jsdelivr.net/npm/[email protected]/dist/medium-zoom.min.js" integrity="sha256-EdPgYcPk/IIrw7FYeuJQexva49pVRZNmt3LculEr7zM=" crossorigin="anonymous"></script>
vendor: 1 bytes, line 1068
1068
1069<script src="/assets/js/zoom.js"></script>
1069 1070 1071 1072<!-- Load Common JS -->
1073<script src="/assets/js/common.js"></script>
1073 1074 1075 1076</html>
Line numbers count LF bytes from the start of the resource, as the search results do. Vendor segments are library code the classifier recognised; they are stored but not indexed. Bytes are shown as Latin1 characters, one per byte.