File size: 50,508 Bytes
0d00d62 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 |
{
"title": "Self-Training Mastery: 100 MCQs",
"description": "A comprehensive set of 100 multiple-choice questions designed to test and deepen your understanding of Self-Training (Semi-Supervised Learning), from fundamental concepts to advanced applications.",
"questions": [
{
"id": 1,
"questionText": "What is the main idea behind self-training in machine learning?",
"options": [
"Train multiple models without using any labeled data",
"Only use unsupervised learning methods",
"Use labeled data to train a model, then iteratively label unlabeled data for further training",
"Reduce dimensionality of the dataset"
],
"correctAnswerIndex": 2,
"explanation": "Self-training uses an initial model trained on labeled data to label unlabeled data. High-confidence predictions are added to the training set iteratively to improve performance."
},
{
"id": 2,
"questionText": "Which type of learning does self-training fall under?",
"options": [
"Reinforcement learning",
"Semi-supervised learning",
"Unsupervised learning only",
"Supervised learning only"
],
"correctAnswerIndex": 1,
"explanation": "Self-training is a semi-supervised learning method because it leverages both labeled and unlabeled data."
},
{
"id": 3,
"questionText": "In self-training, which unlabeled examples are typically added to the training set?",
"options": [
"Only high-confidence predictions from the model",
"All unlabeled examples",
"Randomly selected examples",
"Examples with missing features"
],
"correctAnswerIndex": 0,
"explanation": "Only predictions with high confidence are added to avoid introducing incorrect labels into the training set."
},
{
"id": 4,
"questionText": "What is a key risk in self-training?",
"options": [
"Overfitting due to too much labeled data",
"Lack of model initialization",
"Propagation of incorrect labels",
"Underfitting on labeled data"
],
"correctAnswerIndex": 2,
"explanation": "If the model makes incorrect predictions on unlabeled data and these are added to training, errors can compound over iterations."
},
{
"id": 5,
"questionText": "Which model is typically used as the base in self-training?",
"options": [
"Only neural networks",
"Any supervised classifier",
"Only decision trees",
"Only clustering algorithms"
],
"correctAnswerIndex": 1,
"explanation": "Self-training can use any supervised classifier (e.g., SVM, decision tree, logistic regression) as the initial model."
},
{
"id": 6,
"questionText": "How does self-training differ from co-training?",
"options": [
"Self-training is fully supervised; co-training is unsupervised",
"Co-training uses multiple models trained on different views of data; self-training uses one model",
"There is no difference",
"Self-training uses multiple views; co-training uses a single model"
],
"correctAnswerIndex": 1,
"explanation": "Co-training leverages multiple feature sets (views) and trains separate models, whereas self-training relies on a single model iteratively labeling unlabeled data."
},
{
"id": 7,
"questionText": "Which confidence threshold is important in self-training?",
"options": [
"The learning rate of the optimizer",
"The number of layers in the model",
"The minimum probability required for adding unlabeled predictions to training",
"The batch size"
],
"correctAnswerIndex": 2,
"explanation": "Only predictions exceeding a confidence threshold are considered reliable enough to add to the training set."
},
{
"id": 8,
"questionText": "Self-training is most useful when:",
"options": [
"Labeled data is abundant",
"Only unsupervised tasks are performed",
"There is a small labeled dataset and a large unlabeled dataset",
"Unlabeled data is irrelevant"
],
"correctAnswerIndex": 2,
"explanation": "Self-training leverages the large amount of unlabeled data to improve a model initially trained on limited labeled data."
},
{
"id": 9,
"questionText": "Which step is repeated in self-training iterations?",
"options": [
"Normalize all data repeatedly",
"Shuffle the dataset only",
"Predict unlabeled data and add high-confidence predictions to training set",
"Reduce feature dimensions only"
],
"correctAnswerIndex": 2,
"explanation": "The model predicts labels for unlabeled data and adds confident predictions to expand the labeled set iteratively."
},
{
"id": 10,
"questionText": "What is the primary goal of self-training?",
"options": [
"Remove redundant features",
"Improve model performance using unlabeled data",
"Reduce model complexity",
"Cluster unlabeled data without labels"
],
"correctAnswerIndex": 1,
"explanation": "Self-training aims to improve the model by leveraging additional unlabeled data to provide more training examples."
},
{
"id": 11,
"questionText": "Which metric helps evaluate self-training effectiveness?",
"options": [
"Only confidence scores",
"Batch size",
"Accuracy on a held-out labeled set",
"Number of unlabeled samples"
],
"correctAnswerIndex": 2,
"explanation": "Accuracy on a validation or test set evaluates whether adding pseudo-labeled data improves model performance."
},
{
"id": 12,
"questionText": "What is the key assumption behind self-training?",
"options": [
"Labeled data is insufficient to train a model",
"High-confidence predictions are usually correct",
"All models overfit quickly",
"Unlabeled data is always noisy"
],
"correctAnswerIndex": 1,
"explanation": "Self-training assumes the model's high-confidence predictions are reliable enough to be treated as labeled."
},
{
"id": 13,
"questionText": "Which of the following could worsen self-training performance?",
"options": [
"Using cross-validation",
"High learning rate",
"Large batch size",
"Low confidence threshold allowing incorrect labels"
],
"correctAnswerIndex": 3,
"explanation": "Adding incorrectly predicted labels can propagate errors, harming performance."
},
{
"id": 14,
"questionText": "Can self-training be used for regression tasks?",
"options": [
"No, only clustering works",
"Yes, by predicting continuous labels for unlabeled data",
"No, only classification is allowed",
"Yes, but only with neural networks"
],
"correctAnswerIndex": 1,
"explanation": "Self-training can be applied to regression by adding high-confidence predictions as pseudo-labels to the training set."
},
{
"id": 15,
"questionText": "Which type of data is required for self-training?",
"options": [
"Only labeled data",
"Only unlabeled data",
"Some labeled data and additional unlabeled data",
"Only synthetic data"
],
"correctAnswerIndex": 2,
"explanation": "Self-training needs an initial labeled set to train the base model and a larger unlabeled set to iteratively expand training."
},
{
"id": 16,
"questionText": "Which of the following strategies can improve self-training?",
"options": [
"Add all unlabeled data without filtering",
"Use high-quality initial model and careful confidence thresholding",
"Reduce model size excessively",
"Skip training on labeled data"
],
"correctAnswerIndex": 1,
"explanation": "Starting with a strong initial model and using confidence thresholds ensures pseudo-labeled data is reliable."
},
{
"id": 17,
"questionText": "What is pseudo-labeling in self-training?",
"options": [
"Using labels from a different dataset",
"Assigning predicted labels to unlabeled data for further training",
"Generating random labels for data",
"Removing labels from data"
],
"correctAnswerIndex": 1,
"explanation": "Pseudo-labeling uses the modelβs predictions as labels for unlabeled data to augment the training set."
},
{
"id": 18,
"questionText": "Why is self-training considered semi-supervised?",
"options": [
"It uses reinforcement learning signals",
"It only uses unlabeled data",
"It only uses labeled data",
"It uses both labeled and unlabeled data"
],
"correctAnswerIndex": 3,
"explanation": "Self-training leverages a combination of labeled data and pseudo-labeled unlabeled data, classifying it as semi-supervised learning."
},
{
"id": 19,
"questionText": "Which of the following is a common stopping criterion in self-training?",
"options": [
"All unlabeled data labeled incorrectly",
"Batch size exceeds 1,000",
"Training loss reaches zero",
"No improvement on validation set or maximum iterations reached"
],
"correctAnswerIndex": 3,
"explanation": "Training stops when performance stops improving on validation data or a fixed number of iterations is reached."
},
{
"id": 20,
"questionText": "Which problem occurs if the modelβs predictions are inaccurate in self-training?",
"options": [
"Overfitting to labeled data only",
"Gradient explosion",
"Error reinforcement leading to poor performance",
"Batch normalization failure"
],
"correctAnswerIndex": 2,
"explanation": "Incorrect pseudo-labels are treated as true labels, which may propagate errors and degrade model performance."
},
{
"id": 21,
"questionText": "Which of the following is a benefit of self-training?",
"options": [
"Reduce dependence on large labeled datasets",
"Guarantee perfect accuracy",
"Remove need for model initialization",
"Eliminate need for validation set"
],
"correctAnswerIndex": 0,
"explanation": "Self-training allows models to leverage unlabeled data to improve performance when labeled data is limited."
},
{
"id": 22,
"questionText": "What type of models are most suitable for self-training?",
"options": [
"Only deep neural networks",
"Only clustering algorithms",
"Any supervised model capable of confidence estimation",
"Only decision trees"
],
"correctAnswerIndex": 2,
"explanation": "Any classifier that can output confidence scores for predictions can be used in self-training."
},
{
"id": 23,
"questionText": "Which scenario illustrates effective self-training?",
"options": [
"Randomly labeling all data without model",
"Using only unsupervised clustering",
"Small labeled set of medical images, plus large unlabeled set, iteratively improving classifier",
"Large labeled set only with no unlabeled data"
],
"correctAnswerIndex": 2,
"explanation": "Self-training is effective when the initial labeled set is small and a large unlabeled set exists to augment training."
},
{
"id": 24,
"questionText": "Which factor is critical for self-training success?",
"options": [
"Optimizer type only",
"Number of hidden layers only",
"Accuracy and calibration of initial model predictions",
"Batch size only"
],
"correctAnswerIndex": 2,
"explanation": "If the initial model provides poor or uncalibrated predictions, self-training can propagate errors."
},
{
"id": 25,
"questionText": "Which type of confidence threshold is common in self-training?",
"options": [
"Random selection of all samples",
"Probability above a set value (e.g., 0.9)",
"Minimum batch size",
"Only feature importance threshold"
],
"correctAnswerIndex": 1,
"explanation": "Predictions exceeding a set probability threshold are considered reliable enough to include as pseudo-labels."
},
{
"id": 26,
"questionText": "What happens if self-training is applied with an inaccurate initial model?",
"options": [
"Unlabeled data is ignored",
"Incorrect pseudo-labels can accumulate and degrade performance",
"Only batch normalization is affected",
"Model immediately becomes perfect"
],
"correctAnswerIndex": 1,
"explanation": "The model may reinforce its own mistakes, causing error propagation and poor performance."
},
{
"id": 27,
"questionText": "Which approach can reduce error propagation in self-training?",
"options": [
"Use random pseudo-labeling",
"Reduce latent space dimension",
"Skip model retraining",
"High confidence threshold and careful validation"
],
"correctAnswerIndex": 3,
"explanation": "Filtering pseudo-labels based on high confidence reduces the risk of introducing incorrect labels."
},
{
"id": 28,
"questionText": "Which scenario is not suitable for self-training?",
"options": [
"Small labeled dataset with large unlabeled set",
"When labeled data is large and sufficient",
"Initial model with good accuracy",
"Tasks with uncertainty in unlabeled data"
],
"correctAnswerIndex": 1,
"explanation": "If labeled data is sufficient, semi-supervised methods like self-training offer little benefit."
},
{
"id": 29,
"questionText": "Which evaluation ensures self-training improved the model?",
"options": [
"Count number of pseudo-labeled samples",
"Measure training time only",
"Check confidence scores only",
"Compare performance on held-out labeled validation set before and after iterations"
],
"correctAnswerIndex": 3,
"explanation": "Validation performance demonstrates whether pseudo-labels improved generalization."
},
{
"id": 30,
"questionText": "Which real-world application commonly uses self-training?",
"options": [
"Clustering unlabeled customer data only",
"Large fully labeled image dataset",
"Dimensionality reduction for PCA",
"Text classification with few labeled examples and many unlabeled documents"
],
"correctAnswerIndex": 3,
"explanation": "Self-training is used when labels are scarce, e.g., semi-supervised text classification or medical imaging."
},
{
"id": 31,
"questionText": "Which iterative step is essential in self-training?",
"options": [
"Normalizing all data each iteration",
"Shuffling features only",
"Adding noise to labeled data only",
"Retraining the model with newly pseudo-labeled data"
],
"correctAnswerIndex": 3,
"explanation": "After pseudo-labeling, the model must be retrained on the augmented training set to improve performance."
},
{
"id": 32,
"questionText": "Why is model calibration important in self-training?",
"options": [
"Reduces batch size",
"Removes overfitting completely",
"Speeds up convergence",
"Ensures predicted probabilities reflect true confidence for reliable pseudo-labeling"
],
"correctAnswerIndex": 3,
"explanation": "If confidence scores are miscalibrated, low-quality predictions may be added, causing error propagation."
},
{
"id": 33,
"questionText": "What is a common method to select unlabeled samples in self-training?",
"options": [
"Select the first N samples",
"Pick predictions above a confidence threshold (e.g., 0.9)",
"Select samples with missing features",
"Randomly select samples"
],
"correctAnswerIndex": 1,
"explanation": "High-confidence predictions are more likely to be correct and are therefore safer to add to training."
},
{
"id": 34,
"questionText": "Which model type is less suited for self-training?",
"options": [
"Decision trees",
"Models without probability output or confidence estimation",
"Logistic regression",
"Random forests with probability outputs"
],
"correctAnswerIndex": 1,
"explanation": "Models that cannot estimate prediction confidence cannot safely generate pseudo-labels."
},
{
"id": 35,
"questionText": "What is a typical stopping condition for self-training iterations?",
"options": [
"No significant improvement on validation set",
"Latent space dimensionality stabilizes",
"Training loss reaches zero",
"All unlabeled samples are labeled"
],
"correctAnswerIndex": 0,
"explanation": "Training usually stops when adding more pseudo-labeled samples does not improve validation performance."
},
{
"id": 36,
"questionText": "How can overfitting be minimized in self-training?",
"options": [
"Increase learning rate",
"Use high-quality initial model and validation checks",
"Reduce unlabeled dataset size to zero",
"Use no feature scaling"
],
"correctAnswerIndex": 1,
"explanation": "A reliable initial model and monitoring validation performance prevent overfitting to incorrect pseudo-labels."
},
{
"id": 37,
"questionText": "Which type of confidence threshold is more conservative in self-training?",
"options": [
"No threshold",
"Low threshold (e.g., 0.5)",
"Random threshold",
"High threshold (e.g., 0.95)"
],
"correctAnswerIndex": 3,
"explanation": "High thresholds ensure only very confident predictions are used, reducing error propagation."
},
{
"id": 38,
"questionText": "What is pseudo-label noise?",
"options": [
"Shuffled feature vectors",
"Incorrect labels assigned by the model to unlabeled data",
"Random data augmentation",
"Missing labels in the dataset"
],
"correctAnswerIndex": 1,
"explanation": "Pseudo-label noise refers to incorrectly predicted labels that may be added to the training set, harming performance."
},
{
"id": 39,
"questionText": "In which domain is self-training often applied?",
"options": [
"Fully labeled image datasets only",
"Unsupervised clustering tasks",
"Text classification with limited labeled data",
"Dimensionality reduction tasks"
],
"correctAnswerIndex": 2,
"explanation": "Self-training is commonly used in text classification, NLP, and other domains with limited labeled data."
},
{
"id": 40,
"questionText": "Which variant can improve self-training performance?",
"options": [
"No retraining of the model",
"Random pseudo-label selection",
"Increasing batch size only",
"Entropy-based sample selection"
],
"correctAnswerIndex": 3,
"explanation": "Entropy-based selection chooses samples the model is confident about, improving pseudo-label reliability."
},
{
"id": 41,
"questionText": "Why might self-training fail on highly imbalanced datasets?",
"options": [
"Majority class dominates batch normalization",
"The model may label minority class incorrectly, propagating errors",
"Learning rate is too small",
"Optimizer fails to converge"
],
"correctAnswerIndex": 1,
"explanation": "Imbalanced classes can lead to confident but incorrect pseudo-labels for minority classes, degrading performance."
},
{
"id": 42,
"questionText": "Which strategy helps reduce error propagation in self-training?",
"options": [
"Use uncalibrated predictions",
"Reduce labeled dataset size",
"Remove validation set",
"Use ensemble models to generate pseudo-labels"
],
"correctAnswerIndex": 3,
"explanation": "Ensemble predictions can improve confidence reliability and reduce errors in pseudo-labeling."
},
{
"id": 43,
"questionText": "Which step is optional but can improve self-training?",
"options": [
"Randomly reducing batch size",
"Shuffling features repeatedly",
"Confidence-based filtering or weighting of pseudo-labels",
"Removing labeled data"
],
"correctAnswerIndex": 2,
"explanation": "Weighting or filtering pseudo-labels ensures that unreliable predictions do not harm learning."
},
{
"id": 44,
"questionText": "Which model property is crucial for self-training success?",
"options": [
"Accurate probability/confidence estimates",
"Only large number of hidden layers",
"High batch size only",
"Gradient clipping"
],
"correctAnswerIndex": 0,
"explanation": "Confidence estimates determine which predictions can safely become pseudo-labels."
},
{
"id": 45,
"questionText": "Why is validation important during self-training?",
"options": [
"To shuffle features",
"To calculate batch size",
"To detect whether adding pseudo-labeled data improves generalization",
"To normalize inputs repeatedly"
],
"correctAnswerIndex": 2,
"explanation": "Validation ensures that the iterative pseudo-labeling process actually improves performance."
},
{
"id": 46,
"questionText": "Which sampling method reduces bias in self-training?",
"options": [
"Shuffling labels",
"Selecting only first N samples",
"Selecting diverse or representative unlabeled examples",
"Random batch selection only"
],
"correctAnswerIndex": 2,
"explanation": "Choosing diverse examples helps the model learn from various regions of feature space."
},
{
"id": 47,
"questionText": "Which scenario could cause self-training to stagnate?",
"options": [
"All labeled data is perfect",
"Latent space is too large",
"Batch size is too large",
"No unlabeled data meets confidence threshold"
],
"correctAnswerIndex": 3,
"explanation": "If no unlabeled sample exceeds the confidence threshold, no new pseudo-labels are added, stopping progress."
},
{
"id": 48,
"questionText": "What is co-training, in comparison to self-training?",
"options": [
"Uses multiple models on different feature sets instead of a single model",
"Does not use unlabeled data",
"Uses only a single model",
"Only uses unsupervised clustering"
],
"correctAnswerIndex": 0,
"explanation": "Co-training exploits multiple views of the data to iteratively label unlabeled examples with separate models."
},
{
"id": 49,
"questionText": "Why is early stopping useful in self-training?",
"options": [
"Prevents overfitting to incorrect pseudo-labels",
"Speeds up normalization",
"Reduces batch size",
"Removes labeled data"
],
"correctAnswerIndex": 0,
"explanation": "Stopping training when validation performance plateaus prevents overfitting to potentially noisy pseudo-labels."
},
{
"id": 50,
"questionText": "Which scenario illustrates effective pseudo-label filtering?",
"options": [
"Randomly discard labeled data",
"Select samples with prediction probability > 0.95 for retraining",
"Add all pseudo-labels without checking confidence",
"Use unlabeled data without training"
],
"correctAnswerIndex": 1,
"explanation": "Filtering ensures that only reliable pseudo-labels are used to improve model accuracy."
},
{
"id": 51,
"questionText": "Which factor may reduce self-training effectiveness?",
"options": [
"Balanced labeled data",
"Highly noisy initial model predictions",
"Large unlabeled dataset",
"High-confidence thresholding"
],
"correctAnswerIndex": 1,
"explanation": "A poor initial model can assign wrong pseudo-labels, causing error propagation."
},
{
"id": 52,
"questionText": "What is a practical use case of self-training in NLP?",
"options": [
"Dimensionality reduction of word embeddings",
"Fully supervised large dataset classification",
"Clustering documents without labels",
"Classifying sentiment in a small labeled corpus with many unlabeled reviews"
],
"correctAnswerIndex": 3,
"explanation": "Self-training is useful when labeled text data is limited, but many unlabeled documents are available."
},
{
"id": 53,
"questionText": "Which approach improves pseudo-label reliability?",
"options": [
"Reducing labeled data",
"No retraining",
"Random pseudo-labeling",
"Ensemble predictions from multiple models"
],
"correctAnswerIndex": 3,
"explanation": "Ensemble methods average predictions and reduce error in pseudo-labeling."
},
{
"id": 54,
"questionText": "How can self-training handle class imbalance?",
"options": [
"Use the same threshold for all classes",
"Apply class-specific confidence thresholds or weighted sampling",
"Ignore minority class",
"Remove majority class"
],
"correctAnswerIndex": 1,
"explanation": "Adjusting thresholds or weights prevents minority class samples from being underrepresented in pseudo-labeling."
},
{
"id": 55,
"questionText": "Which type of model output is necessary for self-training?",
"options": [
"Only feature importance",
"Probability or confidence estimates",
"Loss values only",
"Only predicted labels"
],
"correctAnswerIndex": 1,
"explanation": "Confidence scores help determine which pseudo-labels are reliable enough to include."
},
{
"id": 56,
"questionText": "Why might low-confidence pseudo-labels be harmful?",
"options": [
"They improve convergence",
"They may be incorrect and propagate errors",
"They reduce training time",
"They increase batch size unnecessarily"
],
"correctAnswerIndex": 1,
"explanation": "Low-confidence predictions are more likely to be wrong, harming model performance if added."
},
{
"id": 57,
"questionText": "Which technique can enhance self-training robustness?",
"options": [
"Increase batch size only",
"Use dropout, regularization, or ensembles to reduce errors",
"Reduce feature dimensionality only",
"Remove labeled data"
],
"correctAnswerIndex": 1,
"explanation": "Regularization and ensemble techniques reduce overfitting and mitigate error propagation in pseudo-labeling."
},
{
"id": 58,
"questionText": "How can iterative self-training improve a model?",
"options": [
"Remove validation set",
"Gradually expand labeled data with reliable pseudo-labels",
"Use random unlabeled data only",
"Reduce feature dimension only"
],
"correctAnswerIndex": 1,
"explanation": "Iterative addition of confident pseudo-labels allows the model to learn from more data and generalize better."
},
{
"id": 59,
"questionText": "Which factor determines the number of self-training iterations?",
"options": [
"Optimizer type only",
"Convergence of validation performance or exhaustion of high-confidence samples",
"Latent dimension size only",
"Fixed batch size only"
],
"correctAnswerIndex": 1,
"explanation": "Iterations continue until adding pseudo-labels no longer improves validation accuracy or no more confident predictions remain."
},
{
"id": 60,
"questionText": "Which evaluation helps detect pseudo-labeling issues?",
"options": [
"Batch size check",
"Validation accuracy compared to baseline",
"Training loss only",
"Only pseudo-label count"
],
"correctAnswerIndex": 1,
"explanation": "Validation performance reflects whether pseudo-labels actually improve the model."
},
{
"id": 61,
"questionText": "What is the effect of too low a confidence threshold?",
"options": [
"Model training stops immediately",
"Reduces batch size",
"Many incorrect pseudo-labels are added, reducing model performance",
"Latent space shrinks"
],
"correctAnswerIndex": 2,
"explanation": "Low threshold allows unreliable predictions to enter the training set, causing errors."
},
{
"id": 62,
"questionText": "Which method ensures balanced class representation in self-training?",
"options": [
"Use majority class only",
"Remove minority class",
"Class-specific thresholds or weighted sampling",
"Random selection"
],
"correctAnswerIndex": 2,
"explanation": "Adjusting thresholds ensures pseudo-labels for minority classes are included to prevent imbalance."
},
{
"id": 63,
"questionText": "Why is the initial labeled set important?",
"options": [
"It determines batch size",
"It prevents overfitting completely",
"It provides the starting model to generate reliable pseudo-labels",
"It eliminates the need for validation"
],
"correctAnswerIndex": 2,
"explanation": "A good initial model is crucial because errors at this stage can propagate in subsequent iterations."
},
{
"id": 64,
"questionText": "Which step is repeated each iteration in self-training?",
"options": [
"Predict unlabeled data, filter high-confidence predictions, retrain model",
"Shuffle labeled data only",
"Reduce learning rate only",
"Normalize features repeatedly"
],
"correctAnswerIndex": 0,
"explanation": "The core iterative process is labeling, filtering, and retraining."
},
{
"id": 65,
"questionText": "Which metric helps monitor error propagation?",
"options": [
"Batch size",
"Number of unlabeled samples",
"Learning rate",
"Validation accuracy or F1-score"
],
"correctAnswerIndex": 3,
"explanation": "Validation metrics indicate whether pseudo-labels are improving or harming performance."
},
{
"id": 66,
"questionText": "Which technique can mitigate self-training risks?",
"options": [
"No retraining",
"Random pseudo-labeling",
"Ensemble predictions and high confidence thresholds",
"Reduce labeled data"
],
"correctAnswerIndex": 2,
"explanation": "Using ensemble models and filtering reduces error propagation from unreliable pseudo-labels."
},
{
"id": 67,
"questionText": "Which factor can accelerate self-training convergence?",
"options": [
"Random pseudo-labeling",
"Good initial model and large unlabeled dataset",
"Feature shuffling only",
"Small batch size only"
],
"correctAnswerIndex": 1,
"explanation": "A strong initial model provides accurate pseudo-labels, speeding iterative learning."
},
{
"id": 68,
"questionText": "Which scenario illustrates a potential pitfall?",
"options": [
"High-confidence predictions improve model",
"Adding low-confidence predictions leads to compounded errors",
"Validation accuracy increases steadily",
"Ensemble pseudo-labeling reduces errors"
],
"correctAnswerIndex": 1,
"explanation": "Including uncertain predictions can propagate errors, reducing overall performance."
},
{
"id": 69,
"questionText": "Which workflow enhances self-training performance?",
"options": [
"Randomly label all unlabeled data and stop",
"Normalize all features repeatedly without retraining",
"Remove labeled set and retrain",
"Iteratively add high-confidence pseudo-labels, retrain model, validate performance"
],
"correctAnswerIndex": 3,
"explanation": "Following the iterative pseudo-labeling and retraining process ensures learning from unlabeled data."
},
{
"id": 70,
"questionText": "Which evaluation demonstrates self-training success?",
"options": [
"Batch size stability",
"Number of pseudo-labels only",
"Feature normalization success",
"Improved accuracy/F1-score on validation set after iterations"
],
"correctAnswerIndex": 3,
"explanation": "Validation metrics reflect whether the iterative process has improved generalization."
},
{
"id": 71,
"questionText": "In a medical image classification task with limited labeled data, which self-training strategy can minimize error propagation?",
"options": [
"Use an ensemble of classifiers for pseudo-labeling and a high-confidence threshold",
"Add all pseudo-labels regardless of confidence",
"Use only a single weak model without validation",
"Randomly discard labeled examples"
],
"correctAnswerIndex": 0,
"explanation": "Ensembles reduce individual model errors, and a high-confidence threshold ensures only reliable pseudo-labels are added."
},
{
"id": 72,
"questionText": "If a self-training model stagnates, which action is most appropriate?",
"options": [
"Increase learning rate dramatically",
"Stop using pseudo-labeling entirely",
"Lower the confidence threshold carefully to add more unlabeled samples",
"Remove the validation set"
],
"correctAnswerIndex": 2,
"explanation": "Lowering the threshold slightly allows more pseudo-labels to be added while monitoring validation performance."
},
{
"id": 73,
"questionText": "A self-training classifier repeatedly mislabels minority class samples. What is a robust solution?",
"options": [
"Ignore minority class pseudo-labels",
"Use class-specific confidence thresholds or weighted pseudo-label selection",
"Reduce majority class size only",
"Use same threshold for all classes"
],
"correctAnswerIndex": 1,
"explanation": "Adjusting thresholds or weights ensures minority class samples are included reliably in pseudo-labels."
},
{
"id": 74,
"questionText": "During self-training, validation accuracy decreases after adding pseudo-labeled samples. What is the likely cause?",
"options": [
"Error propagation from incorrect pseudo-labels",
"Batch size too small",
"Validation set too large",
"Model has too many layers"
],
"correctAnswerIndex": 0,
"explanation": "Adding incorrect pseudo-labels can harm the model, lowering validation accuracy."
},
{
"id": 75,
"questionText": "Which scenario best illustrates safe pseudo-labeling?",
"options": [
"Using unlabeled data without retraining",
"Adding only predictions above 0.95 confidence and retraining iteratively",
"Randomly adding half of the unlabeled samples",
"Ignoring confidence scores entirely"
],
"correctAnswerIndex": 1,
"explanation": "High-confidence predictions reduce the risk of introducing errors, ensuring reliable iterative learning."
},
{
"id": 76,
"questionText": "In a text classification task, adding pseudo-labels with low confidence caused model collapse. What is the most appropriate adjustment?",
"options": [
"Add more unlabeled samples randomly",
"Shuffle features only",
"Reduce labeled dataset size",
"Increase confidence threshold and monitor validation metrics"
],
"correctAnswerIndex": 3,
"explanation": "Increasing the threshold filters out unreliable pseudo-labels, preventing error accumulation."
},
{
"id": 77,
"questionText": "Which measure can detect pseudo-label noise early?",
"options": [
"Monitoring training loss only",
"Counting pseudo-labels added",
"Monitoring validation accuracy or F1-score after each iteration",
"Measuring batch size"
],
"correctAnswerIndex": 2,
"explanation": "Validation metrics reflect whether pseudo-labels are improving or harming model performance."
},
{
"id": 78,
"questionText": "You have multiple classifiers available. Which strategy enhances self-training reliability?",
"options": [
"Use the weakest classifier only",
"Randomly assign pseudo-labels",
"Ignore labeled dataset entirely",
"Use ensemble predictions for pseudo-labeling"
],
"correctAnswerIndex": 3,
"explanation": "Ensemble predictions reduce individual model errors and improve pseudo-label reliability."
},
{
"id": 79,
"questionText": "If the initial model is poorly calibrated, which self-training risk increases?",
"options": [
"Batch size becomes irrelevant",
"Validation accuracy improves",
"Feature normalization fails",
"High likelihood of adding incorrect pseudo-labels, propagating errors"
],
"correctAnswerIndex": 3,
"explanation": "Poorly calibrated models may produce confident but incorrect predictions, leading to error propagation."
},
{
"id": 80,
"questionText": "Which strategy mitigates imbalance in pseudo-labeling?",
"options": [
"Ignore minority class",
"Use random selection only",
"Remove majority class pseudo-labels entirely",
"Use class-dependent thresholds or weighted sampling"
],
"correctAnswerIndex": 3,
"explanation": "Adjusting thresholds ensures minority classes are represented in the pseudo-labeled training set."
},
{
"id": 81,
"questionText": "Scenario: A self-training model improves initially but then validation performance plateaus. Recommended action?",
"options": [
"Add random pseudo-labels",
"Increase model depth without bound",
"Reduce labeled dataset size",
"Stop iterating or consider adjusting confidence threshold"
],
"correctAnswerIndex": 3,
"explanation": "Iteration should stop when additional pseudo-labeling no longer improves validation accuracy."
},
{
"id": 82,
"questionText": "Which approach reduces the chance of catastrophic error propagation in self-training?",
"options": [
"Ignore validation metrics",
"Add pseudo-labels indiscriminately",
"Combine ensemble models and high-confidence thresholds",
"Randomly discard labeled samples"
],
"correctAnswerIndex": 2,
"explanation": "Ensemble models and filtering based on confidence help ensure that only reliable pseudo-labels are used."
},
{
"id": 83,
"questionText": "In active learning combined with self-training, which samples are prioritized for human labeling?",
"options": [
"High-confidence pseudo-labels",
"Low-confidence predictions or uncertain samples",
"Previously labeled samples",
"Random unlabeled samples"
],
"correctAnswerIndex": 1,
"explanation": "Human labeling focuses on uncertain samples to improve model learning and reduce pseudo-label errors."
},
{
"id": 84,
"questionText": "You notice class imbalance increases after adding pseudo-labels. Which solution is appropriate?",
"options": [
"Adjust pseudo-label selection thresholds per class",
"Randomly add more pseudo-labels",
"Ignore imbalance",
"Remove minority class"
],
"correctAnswerIndex": 0,
"explanation": "Class-specific thresholds ensure balanced representation and prevent bias toward majority classes."
},
{
"id": 85,
"questionText": "In image classification, which self-training design reduces error propagation?",
"options": [
"Ignore unlabeled data",
"Combine ensemble classifiers, apply high-confidence threshold, validate iteratively",
"Add all pseudo-labels without validation",
"Randomly relabel labeled data"
],
"correctAnswerIndex": 1,
"explanation": "This design ensures only reliable pseudo-labels are added and iteratively validated."
},
{
"id": 86,
"questionText": "Scenario: Model performs well on majority class but poorly on minority class after pseudo-labeling. Best remedy?",
"options": [
"Increase batch size only",
"Ignore minority class",
"Use class-weighted thresholds or reweight loss function",
"Remove validation set"
],
"correctAnswerIndex": 2,
"explanation": "Weighted thresholds or loss functions help the model learn effectively across all classes."
},
{
"id": 87,
"questionText": "Which metric is more informative for self-training in imbalanced datasets?",
"options": [
"F1-score or class-specific precision/recall",
"Number of pseudo-labeled samples",
"Batch size",
"Overall accuracy only"
],
"correctAnswerIndex": 0,
"explanation": "Metrics considering class balance provide better insight into pseudo-labeling performance."
},
{
"id": 88,
"questionText": "Which step prevents pseudo-labeling from reinforcing errors?",
"options": [
"Use uncalibrated predictions",
"Add all pseudo-labels indiscriminately",
"Filter predictions with confidence above threshold and validate",
"Ignore validation metrics"
],
"correctAnswerIndex": 2,
"explanation": "Filtering high-confidence predictions and validating ensures pseudo-labels are reliable."
},
{
"id": 89,
"questionText": "Which approach combines self-training and semi-supervised consistency regularization?",
"options": [
"Add pseudo-labels randomly",
"Use model predictions on augmented unlabeled samples to enforce consistent outputs",
"Only retrain on labeled data",
"Ignore data augmentation"
],
"correctAnswerIndex": 1,
"explanation": "Consistency regularization encourages the model to produce stable predictions across perturbations, improving pseudo-label reliability."
},
{
"id": 90,
"questionText": "Scenario: Large unlabeled dataset, initial model low accuracy. Self-training fails. Best strategy?",
"options": [
"Remove labeled data",
"Add all unlabeled data randomly",
"Improve initial model, apply ensemble pseudo-labeling, high-confidence filtering",
"Reduce dataset dimensionality only"
],
"correctAnswerIndex": 2,
"explanation": "A better initial model reduces error propagation and improves iterative self-training outcomes."
},
{
"id": 91,
"questionText": "Which strategy can prevent majority-class dominance in pseudo-labeling?",
"options": [
"Increase batch size only",
"Use uniform threshold",
"Apply class-dependent thresholds and sample weighting",
"Ignore minority class"
],
"correctAnswerIndex": 2,
"explanation": "Class-specific thresholds prevent pseudo-labeling from overrepresenting majority class."
},
{
"id": 92,
"questionText": "How can pseudo-label reliability be evaluated before retraining?",
"options": [
"Measure batch size only",
"Check agreement between ensemble predictions and existing labeled data",
"Count number of pseudo-labels",
"Monitor feature variance"
],
"correctAnswerIndex": 1,
"explanation": "Agreement among multiple models indicates reliability of pseudo-labels."
},
{
"id": 93,
"questionText": "Scenario: Self-training on text data with domain shift. What adjustment is helpful?",
"options": [
"Use domain-adapted model or adjust pseudo-label selection based on domain confidence",
"Add pseudo-labels indiscriminately",
"Ignore domain differences",
"Remove unlabeled data"
],
"correctAnswerIndex": 0,
"explanation": "Domain adaptation ensures pseudo-labels remain reliable even if the unlabeled data distribution differs."
},
{
"id": 94,
"questionText": "Which workflow ensures safe iterative self-training?",
"options": [
"Predict β Add all unlabeled β Retrain",
"Only train on labeled data",
"Predict β Filter high-confidence β Retrain β Validate β Repeat",
"Ignore pseudo-label confidence"
],
"correctAnswerIndex": 2,
"explanation": "This iterative process ensures pseudo-labels improve generalization without propagating errors."
},
{
"id": 95,
"questionText": "Scenario: Self-training on medical images. Some pseudo-labels are noisy. Best mitigation?",
"options": [
"Add all pseudo-labels without check",
"Ignore labeled dataset",
"Reduce batch size only",
"Use model ensembles, high-confidence filtering, validate iteratively"
],
"correctAnswerIndex": 3,
"explanation": "Combining ensembles and confidence-based filtering reduces noise impact."
},
{
"id": 96,
"questionText": "Which strategy can handle evolving data distributions in self-training?",
"options": [
"Iteratively re-evaluate pseudo-label confidence and adjust thresholds",
"Remove labeled data",
"Fix threshold for all iterations",
"Add pseudo-labels blindly"
],
"correctAnswerIndex": 0,
"explanation": "Adapting thresholds ensures pseudo-labeling remains reliable as data distribution changes."
},
{
"id": 97,
"questionText": "Scenario: Model collapses after multiple self-training iterations. Primary cause?",
"options": [
"Accumulation of incorrect pseudo-labels and error propagation",
"Learning rate too high",
"Feature scaling failed",
"Batch size too small"
],
"correctAnswerIndex": 0,
"explanation": "Incorrect pseudo-labels added iteratively can compound errors and degrade model performance."
},
{
"id": 98,
"questionText": "Which approach improves self-training robustness on small labeled datasets?",
"options": [
"Ignore validation set",
"Add pseudo-labels randomly",
"Start with strong initial model, use ensembles, high-confidence thresholds, and validation",
"Reduce batch size excessively"
],
"correctAnswerIndex": 2,
"explanation": "Careful initial model selection, ensemble predictions, and validation reduce risk of error propagation."
},
{
"id": 99,
"questionText": "Scenario: Self-training applied to regression. Which method ensures reliability?",
"options": [
"Add all pseudo-labels regardless of predicted variance",
"Add only pseudo-labels where predicted values have low uncertainty",
"Ignore pseudo-labels",
"Shuffle labels randomly"
],
"correctAnswerIndex": 1,
"explanation": "Filtering pseudo-labels based on prediction confidence or variance ensures reliable iterative learning."
},
{
"id": 100,
"questionText": "Which combination best prevents failure in large-scale self-training?",
"options": [
"Ignore validation metrics",
"Single weak model with random pseudo-labels",
"Ensemble models, high-confidence filtering, validation checks, iterative retraining",
"Add all pseudo-labels regardless of confidence"
],
"correctAnswerIndex": 2,
"explanation": "This comprehensive strategy ensures pseudo-labels are reliable and model performance improves safely."
}
]
}
|