| [ |
| { |
| "question_id": "Q0001", |
| "paper_id": "2024MNRAS.535.1844L", |
| "question": "How can one obtain accurate photometric redshift estimates for high-redshift galaxies using only flux measurements from different observational filters as input?", |
| "option_a": "MLP", |
| "option_b": "LSTM", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0002", |
| "paper_id": "2024ApJ...977...19D", |
| "question": "How can the global 21-cm signal be reliably modeled from a specified set of cosmological and astrophysical parameters?", |
| "option_a": "Two-layer LSTM", |
| "option_b": "VAE", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0003", |
| "paper_id": "2024ApJ...977...19D", |
| "question": "How can the global 21-cm signal be reliably modeled from a specified set of cosmological and astrophysical parameters?", |
| "option_a": "Two-layer LSTM", |
| "option_b": "Single MLP", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0004", |
| "paper_id": "2024ApJ...977...19D", |
| "question": "How can the global 21-cm signal be reliably modeled from a specified set of cosmological and astrophysical parameters?", |
| "option_a": "Two-layer LSTM", |
| "option_b": "Multi-MLP + PCA + Decision Tree", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0005", |
| "paper_id": "2024AJ....168..242A", |
| "question": "How can the near-surface gravity field of an irregular, nonuniform small body be accurately reconstructed from sampled simulated gravity-field data?", |
| "option_a": "MLP", |
| "option_b": "Neural density field", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0006", |
| "paper_id": "2024AJ....168..242A", |
| "question": "How can the near-surface gravity field of an irregular, nonuniform small body be accurately reconstructed from sampled simulated gravity-field data?", |
| "option_a": "vanilla PINN", |
| "option_b": "Neural density field", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0007", |
| "paper_id": "2024AJ....168..242A", |
| "question": "How can the near-surface gravity field of an irregular, nonuniform small body be accurately reconstructed from sampled simulated gravity-field data?", |
| "option_a": "Transformer-PINN", |
| "option_b": "Neural density field", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0008", |
| "paper_id": "2024AJ....168..242A", |
| "question": "How can the near-surface gravity field of an irregular, nonuniform small body be accurately reconstructed from sampled simulated gravity-field data?", |
| "option_a": "Neural density field", |
| "option_b": "Self-adaptive PINN", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0009", |
| "paper_id": "2024AJ....168..242A", |
| "question": "How can the near-surface gravity field of an irregular, nonuniform small body be accurately reconstructed from sampled simulated gravity-field data?", |
| "option_a": "Self-adaptive PINN", |
| "option_b": "MLP", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0010", |
| "paper_id": "2024AJ....168..242A", |
| "question": "How can the near-surface gravity field of an irregular, nonuniform small body be accurately reconstructed from sampled simulated gravity-field data?", |
| "option_a": "Self-adaptive PINN", |
| "option_b": "vanilla PINN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0011", |
| "paper_id": "2024AJ....168..242A", |
| "question": "How can the near-surface gravity field of an irregular, nonuniform small body be accurately reconstructed from sampled simulated gravity-field data?", |
| "option_a": "Self-adaptive PINN", |
| "option_b": "Transformer-PINN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0012", |
| "paper_id": "2024AJ....168..242A", |
| "question": "How can the near-surface gravity field of an irregular, nonuniform small body be accurately reconstructed from sampled simulated gravity-field data?", |
| "option_a": "MLP", |
| "option_b": "Transformer-PINN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0013", |
| "paper_id": "2024AJ....168..242A", |
| "question": "How can the near-surface gravity field of an irregular, nonuniform small body be accurately reconstructed from sampled simulated gravity-field data?", |
| "option_a": "Transformer-PINN", |
| "option_b": "vanilla PINN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0014", |
| "paper_id": "2024AJ....168..242A", |
| "question": "How can the near-surface gravity field of an irregular, nonuniform small body be accurately reconstructed from sampled simulated gravity-field data?", |
| "option_a": "vanilla PINN", |
| "option_b": "MLP", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0015", |
| "paper_id": "2024A&A...692A.132H", |
| "question": "How can we accurately and precisely measure time delays from simulated microlensed light curves of strongly lensed Type Ia supernovae?", |
| "option_a": "Random Forest", |
| "option_b": "LSTM + fully connected neural network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0016", |
| "paper_id": "2024A&A...692A..72C", |
| "question": "For identifying galaxy-scale strong lenses in multiband survey images, which CNN architecture gives higher recall with very few false positives?", |
| "option_a": "CNN", |
| "option_b": "ResNet", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0017", |
| "paper_id": "2024A&A...692A..72C", |
| "question": "For identifying galaxy-scale strong lenses in multiband survey images, which CNN architecture gives higher recall with very few false positives?", |
| "option_a": "Group-equivariant CNN", |
| "option_b": "CNN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0018", |
| "paper_id": "2024A&A...692A..72C", |
| "question": "How can we identify galaxy-scale strong gravitational lenses in multi-band survey images while maximizing recall at zero or very low contamination?", |
| "option_a": "Group-equivariant CNN", |
| "option_b": "ResNet", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0019", |
| "paper_id": "2024ApJ...976..143T", |
| "question": "For quasar spectra, how can the unabsorbed continuum in the Lyα forest be more accurately predicted from only spectral information redward of Lyα?", |
| "option_a": "PCA", |
| "option_b": "CNN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0020", |
| "paper_id": "2024ApJ...976..143T", |
| "question": "For quasar spectra, how can the unabsorbed continuum in the Lyα forest be more accurately predicted from only spectral information redward of Lyα?", |
| "option_a": "CNN", |
| "option_b": "Non-Negative Matrix Factorization", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0021", |
| "paper_id": "2024ApJ...976...37W", |
| "question": "In cosmological hydrodynamical simulations, how can galaxy stellar masses be more accurately predicted from dark-matter subhalo properties and large-scale environment?", |
| "option_a": "Explainable Boosting Machines", |
| "option_b": "E(3)-invariant GNN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0022", |
| "paper_id": "2024A&A...691A.338C", |
| "question": "For Solar System survey tracklets, how can interstellar objects be more reliably distinguished from other Solar System objects while minimizing false positives and false negatives?", |
| "option_a": "Neural Network", |
| "option_b": "Gradient Boosting Machines", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0023", |
| "paper_id": "2024A&A...691A.338C", |
| "question": "For Solar System survey tracklets, how can interstellar objects be more reliably distinguished from other Solar System objects while minimizing false positives and false negatives?", |
| "option_a": "Random Forest", |
| "option_b": "Neural Network", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0024", |
| "paper_id": "2024A&A...691A.223V", |
| "question": "Using low-resolution optical spectra, how can hot subdwarf binaries be more accurately identified against single hot subdwarfs?", |
| "option_a": "CNN", |
| "option_b": "Self-Organizing Map", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0025", |
| "paper_id": "2024A&A...691A.223V", |
| "question": "Using optical photometry and low-resolution spectrophotometric observations, which approach more accurately identifies hot subdwarf binaries against single hot subdwarfs?", |
| "option_a": "weighted SVM", |
| "option_b": "CNN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0026", |
| "paper_id": "2024ApJ...974..159J", |
| "question": "Using five-band galaxy images and photometry, how can photometric redshifts be estimated more reliably for weak-lensing cosmology?", |
| "option_a": "Bayesian CNN", |
| "option_b": "XGBoost", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0027", |
| "paper_id": "2024ApJ...974..159J", |
| "question": "Using five-band galaxy images and photometry, how can photometric redshifts be estimated more reliably for weak-lensing cosmology?", |
| "option_a": "Bayesian CNN", |
| "option_b": "Bayesian MLP", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0028", |
| "paper_id": "2024ApJ...974..159J", |
| "question": "Using five-band galaxy images and photometry, how can photometric redshifts be estimated more reliably for weak-lensing cosmology?", |
| "option_a": "Bayesian CNN", |
| "option_b": "MLP", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0029", |
| "paper_id": "2024ApJ...974..159J", |
| "question": "Using five-band galaxy images and photometry, how can photometric redshifts be estimated more reliably for weak-lensing cosmology?", |
| "option_a": "SVM", |
| "option_b": "Bayesian CNN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0030", |
| "paper_id": "2024ApJ...974..159J", |
| "question": "Using five-band galaxy images and photometry, how can photometric redshifts be estimated more reliably for weak-lensing cosmology?", |
| "option_a": "XGBoost", |
| "option_b": "CNN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0031", |
| "paper_id": "2024ApJ...974..159J", |
| "question": "Using five-band galaxy images and photometry, how can photometric redshifts be estimated more reliably for weak-lensing cosmology?", |
| "option_a": "MLP", |
| "option_b": "CNN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0032", |
| "paper_id": "2024ApJ...974..159J", |
| "question": "Using five-band galaxy images and photometry, how can photometric redshifts be estimated more reliably for weak-lensing cosmology?", |
| "option_a": "SVM", |
| "option_b": "CNN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0033", |
| "paper_id": "2024ApJ...974...96S", |
| "question": "For astrometric candidate pairs, how can true astrometric binaries be distinguished from visual alignments with high discriminating power using proper motions, parallaxes, and angular and physical separations?", |
| "option_a": "Logistic regression", |
| "option_b": "Artificial neural network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0034", |
| "paper_id": "2024ApJ...973..163Y", |
| "question": "For compact-binary phase-space trajectories learned from numerical simulations, how can orbital evolution be predicted more accurately and efficiently over longer prediction horizons?", |
| "option_a": "Temporal Convolutional Network", |
| "option_b": "LSTM", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0035", |
| "paper_id": "2024ApJ...973..154H", |
| "question": "How can stellar evolutionary tracks and observables be emulated more accurately from grids of stellar evolution models?", |
| "option_a": "Conditional Normalizing Flows", |
| "option_b": "Nearest-Neighbor Interpolation", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0036", |
| "paper_id": "2024A&A...690A.224L", |
| "question": "How can QSO classifications based on broadband colors achieve higher precision in a joint star-galaxy-QSO classification task?", |
| "option_a": "hierarchical density-based clustering", |
| "option_b": "sharpened-dimensionality-reduction + k-nearest-neighbor classification", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0037", |
| "paper_id": "2024RAA....24i5023F", |
| "question": "For stellar spectra represented as wavelet images, how can spectral types be classified more accurately?", |
| "option_a": "Vision Transformer", |
| "option_b": "feature-pyramid ConvNeXt-V2 CNN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0038", |
| "paper_id": "2024RAA....24i5023F", |
| "question": "For stellar spectra represented as wavelet images, how can spectral types be classified more accurately?", |
| "option_a": "Random Forest", |
| "option_b": "feature-pyramid ConvNeXt-V2 CNN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0039", |
| "paper_id": "2024RAA....24i5023F", |
| "question": "For stellar spectra represented as wavelet images, how can spectral types be classified more accurately?", |
| "option_a": "stochastic-gradient-descent linear classifier", |
| "option_b": "feature-pyramid ConvNeXt-V2 CNN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0040", |
| "paper_id": "2024RAA....24i5022M", |
| "question": "For sunspot groups observed in white-light and magnetogram images, how can their magnetic classes be classified more reliably when edge dimming weakens image features?", |
| "option_a": "dual-stream ResNet50", |
| "option_b": "dual-stream ResNet50 with channel-and-spatial attention", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0041", |
| "paper_id": "2024RAA....24i5022M", |
| "question": "For sunspot groups observed in white-light and magnetogram images, how can their magnetic classes be classified more reliably when edge dimming weakens image features?", |
| "option_a": "dual-stream ResNet50 with squeeze-and-excitation attention", |
| "option_b": "dual-stream ResNet50 with channel-and-spatial attention", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0042", |
| "paper_id": "2024MNRAS.533..525M", |
| "question": "Across combined real-and-simulated strong-lens test sets, which neural-network design gives more stable overall lens-classification performance?", |
| "option_a": "five-layer classical CNN", |
| "option_b": "ResNet-based CNN", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0043", |
| "paper_id": "2024MNRAS.533..525M", |
| "question": "For strong-lens searches evaluated on real lens candidates and real non-lenses, which neural-network design gives a better precision-recall trade-off at the adopted survey-search threshold?", |
| "option_a": "ResNet-based CNN", |
| "option_b": "block-structured CNN with batch normalization", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0044", |
| "paper_id": "2024ApJS..274....7M", |
| "question": "For low-to-high-quality galaxy-image super-resolution, which approach produces more visually realistic galaxy reconstructions?", |
| "option_a": "SRGAN", |
| "option_b": "EDSR", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0045", |
| "paper_id": "2024ApJS..274....7M", |
| "question": "For low-to-high-quality galaxy-image super-resolution, which approach produces more visually realistic galaxy reconstructions?", |
| "option_a": "SRGAN", |
| "option_b": "RCAN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0046", |
| "paper_id": "2024ApJS..274....7M", |
| "question": "For low-to-high-quality galaxy-image super-resolution, which approach produces more visually realistic galaxy reconstructions?", |
| "option_a": "SRGAN", |
| "option_b": "efficient nonlocal contrastive attention network", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0047", |
| "paper_id": "2024ApJS..274....7M", |
| "question": "For low-to-high-quality galaxy-image super-resolution, which approach better matches high-resolution reference images in pixel-level and structural fidelity?", |
| "option_a": "efficient nonlocal contrastive attention network", |
| "option_b": "SRGAN", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0048", |
| "paper_id": "2024ApJ...973...41Z", |
| "question": "For Bayesian inference of reionization parameters from pure-signal 3D 21 cm light-cone images, which summary-compression approach yields tighter posterior constraints?", |
| "option_a": "solid harmonic wavelet scattering transform + DELFI", |
| "option_b": "3D CNN + DELFI", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0049", |
| "paper_id": "2024ApJ...973...41Z", |
| "question": "For Bayesian inference of reionization parameters from 3D 21 cm light-cone images, which summary-compression approach yields tighter posterior constraints?", |
| "option_a": "power spectrum + DELFI", |
| "option_b": "solid harmonic wavelet scattering transform + DELFI", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0050", |
| "paper_id": "2024ApJ...972....7R", |
| "question": "For bright-transient survey candidates, which architecture better supports automated spectroscopic follow-up when source-level completeness and purity are prioritized?", |
| "option_a": "multi-modal CNN", |
| "option_b": "fully connected neural network", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0051", |
| "paper_id": "2024ApJ...972....7R", |
| "question": "For identifying bright extragalactic transients in survey alerts, which ML pipeline gives higher classification accuracy?", |
| "option_a": "CNN-MLP fusion of images with photometric and catalogue features", |
| "option_b": "CNN using images", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0052", |
| "paper_id": "2024ApJ...972....7R", |
| "question": "For classifying individual bright-transient survey alert packets before source-level follow-up policies are applied, which unimodal architecture gives higher alert-level accuracy and ROC-AUC?", |
| "option_a": "fully connected neural network", |
| "option_b": "uni-modal CNN", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0053", |
| "paper_id": "2024A&A...689A.181M", |
| "question": "For mock galaxy-cluster velocity-dispersion maps, which U-Net variant better recovers the radial ICL-fraction structure while reducing spurious small-scale ICL features?", |
| "option_a": "Attention U-Net", |
| "option_b": "U-Net", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0054", |
| "paper_id": "2024A&A...689A.166C", |
| "question": "For emission-line galaxy target selection across the tested photometric configurations, which classifier more consistently maintains higher target precision at high target recall?", |
| "option_a": "Self-Organizing Map", |
| "option_b": "MLP", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0055", |
| "paper_id": "2024A&A...689A.166C", |
| "question": "For emission-line galaxy target selection across the tested photometric configurations, which classifier more consistently maintains higher target precision at high target recall?", |
| "option_a": "Self-Organizing Map", |
| "option_b": "SVM", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0056", |
| "paper_id": "2024A&A...689A.166C", |
| "question": "For emission-line galaxy target selection across the tested photometric configurations, which classifier more consistently maintains higher target precision at high target recall?", |
| "option_a": "Self-Organizing Map", |
| "option_b": "Extra Trees Classifier", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0057", |
| "paper_id": "2024A&A...689A.166C", |
| "question": "For emission-line galaxy target selection across the tested photometric configurations, which classifier more consistently maintains higher target precision at high target recall?", |
| "option_a": "Random Forest", |
| "option_b": "Self-Organizing Map", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0058", |
| "paper_id": "2024A&A...689A.166C", |
| "question": "For emission-line galaxy target selection across the tested photometric configurations, which classifier more consistently maintains higher target precision at high target recall?", |
| "option_a": "AdaBoost", |
| "option_b": "MLP", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0059", |
| "paper_id": "2024A&A...689A.166C", |
| "question": "For emission-line galaxy target selection across the tested photometric configurations, which classifier more consistently maintains higher target precision at high target recall?", |
| "option_a": "SVM", |
| "option_b": "AdaBoost", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0060", |
| "paper_id": "2024A&A...689A.143G", |
| "question": "For faint companion spectra cross-correlated with multiple molecular templates, which classifier detects planets more sensitively at low false discovery rate when atmospheric characteristics are uncertain?", |
| "option_a": "1D CNN", |
| "option_b": "Perceptron", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0061", |
| "paper_id": "2024MNRAS.532.3322A", |
| "question": "For identifying multi-component radio sources, which ML approach gives higher classification accuracy?", |
| "option_a": "CNN using radio images", |
| "option_b": "CNN-ANN fusion of radio images and catalogue features", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0062", |
| "paper_id": "2024MNRAS.532.3322A", |
| "question": "For identifying multi-component radio sources, which ML approach gives higher classification accuracy?", |
| "option_a": "ANN using catalogue features", |
| "option_b": "CNN-ANN fusion of radio images and catalogue features", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0063", |
| "paper_id": "2024ApJS..273...27S", |
| "question": "For reconstructing the cosmic distance ladder from Type Ia supernova apparent magnitudes with full covariance information, which method better predicts distances while preserving a smooth and monotonic redshift-distance relation?", |
| "option_a": "Support Vector Regression", |
| "option_b": "covariance-aware LSTM", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0064", |
| "paper_id": "2024ApJS..273...27S", |
| "question": "For reconstructing the cosmic distance ladder from Type Ia supernova apparent magnitudes with full covariance information, which method better predicts distances while preserving a smooth and monotonic redshift-distance relation?", |
| "option_a": "MLP", |
| "option_b": "covariance-aware LSTM", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0065", |
| "paper_id": "2024ApJS..273...27S", |
| "question": "For reconstructing the cosmic distance ladder from Type Ia supernova apparent magnitudes with full covariance information, which method better predicts distances while preserving a smooth and monotonic redshift-distance relation?", |
| "option_a": "kNN regression", |
| "option_b": "covariance-aware LSTM", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0066", |
| "paper_id": "2024ApJ...970..120D", |
| "question": "For extracting sunspot components from heterogeneous historical sunspot drawings when labels are available from one station and other stations are unlabeled, which segmentation approach gives more accurate cross-station segmentation?", |
| "option_a": "Hybrid Task Cascade R-CNN", |
| "option_b": "semisupervised dynamic mutual training with boundary-guided segmentation", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0067", |
| "paper_id": "2024ApJ...970..120D", |
| "question": "For extracting sunspot components from heterogeneous historical sunspot drawings when labels are available from one station and other stations are unlabeled, which segmentation approach gives more accurate cross-station segmentation?", |
| "option_a": "semisupervised dynamic mutual training with boundary-guided segmentation", |
| "option_b": "scale-aware attention Mask R-CNN with PointRend", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0068", |
| "paper_id": "2024ApJ...970..120D", |
| "question": "For extracting sunspot components from heterogeneous historical sunspot drawings when labels are available from one station and other stations are unlabeled, which segmentation approach gives more accurate cross-station segmentation?", |
| "option_a": "semisupervised dynamic mutual training with boundary-guided segmentation", |
| "option_b": "dynamic mutual training with DeepLab-v3", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0069", |
| "paper_id": "2024A&A...688A.199P", |
| "question": "For compressing cosmic 21 cm lightcones for astrophysical parameter inference across the prior volume, which summary approach yields the tightest typical Fisher-forecast constraints?", |
| "option_a": "2D power spectrum + information-maximizing neural network", |
| "option_b": "RNN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0070", |
| "paper_id": "2024A&A...688A.199P", |
| "question": "Among neural-network summaries of cosmic 21 cm lightcones, which approach gives tighter typical astrophysical parameter constraints across the prior volume?", |
| "option_a": "Information-maximizing neural network", |
| "option_b": "RNN", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0071", |
| "paper_id": "2024MNRAS.531.4363S", |
| "question": "For separating accreted from in-situ stars in Milky Way-like galaxies when a classifier trained on one cosmological simulation suite must generalize to another, which method is less sensitive to domain shift?", |
| "option_a": "MLP", |
| "option_b": "XGBoost", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0072", |
| "paper_id": "2024ApJS..273....3A", |
| "question": "For simulated high-dynamic-range radio-interferometric imaging, how can images be reconstructed with higher SNR/logSNR while suppressing hallucinated structure?", |
| "option_a": "U-Net", |
| "option_b": "residual-to-residual DNN series", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0073", |
| "paper_id": "2024ApJS..273....3A", |
| "question": "For simulated radio-interferometric imaging, which learned reconstruction approach better preserves image-domain data fidelity while maintaining substantially higher reconstruction quality?", |
| "option_a": "U-Net", |
| "option_b": "plug-and-play learned-denoiser reconstruction", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0074", |
| "paper_id": "2024MNRAS.531.1454K", |
| "question": "When assigning optical emission-line equivalent widths to mock galaxy spectra, which empirical mapping is less tied to assumptions from a separate stellar-population model while giving comparable or slightly better EW predictions?", |
| "option_a": "feed-forward neural network", |
| "option_b": "gradient-boosted decision trees", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0075", |
| "paper_id": "2024MNRAS.533.1976M", |
| "question": "For classifying 1D optical spectra into quasars, galaxies, stars, and BAL quasars, which neural-network architecture gives higher classification accuracy when trained on the same spectra?", |
| "option_a": "Transformer", |
| "option_b": "1D residual CNN", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0076", |
| "paper_id": "2024A&A...687A.202W", |
| "question": "For estimating wavefront aberrations from non-modulated pyramid wavefront-sensor images in adaptive optics, which neural-network architecture gives lower wavefront-estimation error?", |
| "option_a": "global-context vision Transformer", |
| "option_b": "ConvNeXt CNN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0077", |
| "paper_id": "2024A&A...687A.202W", |
| "question": "For estimating wavefront aberrations from non-modulated pyramid wavefront-sensor images in adaptive optics, which neural-network architecture gives lower wavefront-estimation error?", |
| "option_a": "wavefront-sensing CNN", |
| "option_b": "global-context vision Transformer", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0078", |
| "paper_id": "2024MNRAS.528.5890P", |
| "question": "For estimating stellar surface gravity from stellar light curves, which neural-network architecture gives lower prediction error?", |
| "option_a": "convolution-augmented Transformer", |
| "option_b": "ResNet CNN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0079", |
| "paper_id": "2024MNRAS.528.5890P", |
| "question": "For estimating stellar surface gravity from stellar light curves, which neural-network architecture gives lower prediction error?", |
| "option_a": "convolution-augmented Transformer", |
| "option_b": "convolution-only network", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0080", |
| "paper_id": "2024MNRAS.528.5890P", |
| "question": "For estimating stellar surface gravity from stellar light curves, which neural-network architecture gives lower prediction error?", |
| "option_a": "convolution-augmented Transformer", |
| "option_b": "self-attention-only Transformer", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0081", |
| "paper_id": "2024MNRAS.528.5890P", |
| "question": "For estimating stellar surface gravity from stellar light curves, which neural-network architecture gives lower prediction error?", |
| "option_a": "convolution-only network", |
| "option_b": "self-attention-only Transformer", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0082", |
| "paper_id": "2024MNRAS.528.5890P", |
| "question": "For estimating stellar surface gravity across the red-giant branch from stellar light-curve information, which machine-learning approach gives lower RMSE and fewer outlier predictions?", |
| "option_a": "Transformer", |
| "option_b": "k-nearest-neighbor regression", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0083", |
| "paper_id": "2024AJ....167..118R", |
| "question": "For compressing Solar System object feature vectors used in anomaly searches, which dimensionality-reduction method preserves the features more accurately?", |
| "option_a": "principal component analysis", |
| "option_b": "deep autoencoder", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0084", |
| "paper_id": "2024AJ....167..118R", |
| "question": "For compressing Solar System object feature vectors used in anomaly searches, which dimensionality-reduction method preserves the features more accurately?", |
| "option_a": "non-negative matrix factorization", |
| "option_b": "deep autoencoder", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0085", |
| "paper_id": "2024ApJS..272...14H", |
| "question": "Within a large survey variable-source classification system, for identifying RR Lyrae ab variable stars from light-curve features, which supervised ML method gives higher precision and recall?", |
| "option_a": "XGBoost gradient-boosted decision trees", |
| "option_b": "convolutional/dense neural network", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0086", |
| "paper_id": "2024ApJS..272...14H", |
| "question": "Within a large survey variable-source classification system, for identifying Delta Scuti variable stars from light-curve features, which supervised ML method gives higher precision and recall?", |
| "option_a": "XGBoost gradient-boosted decision trees", |
| "option_b": "convolutional/dense neural network", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0087", |
| "paper_id": "2024ApJS..272...14H", |
| "question": "Within a large survey variable-source classification system, for identifying active galactic nuclei from light-curve features, which supervised ML method gives higher precision and recall?", |
| "option_a": "convolutional/dense neural network", |
| "option_b": "XGBoost gradient-boosted decision trees", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0088", |
| "paper_id": "2024A&A...681A..68E", |
| "question": "For identifying galaxy-galaxy strong-lensing systems in high-resolution space-like images when less prominent lensing features are included, which CNN architecture gives higher classification accuracy and AUC?", |
| "option_a": "residual CNN", |
| "option_b": "inception CNN", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0089", |
| "paper_id": "2024A&A...692A.260R", |
| "question": "For AGN photometric-redshift estimation in wide-area survey data where host-galaxy and nuclear emission can be spatially mixed, which ML approach gives lower scatter and outlier fraction on an independent blind sample?", |
| "option_a": "image-based CNN ensemble", |
| "option_b": "fully connected neural network using catalog multi-aperture photometry", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0090", |
| "paper_id": "2024AJ....168..234L", |
| "question": "For identifying true stellar flares from flare-candidate profile features, which supervised classifier is more reliable?", |
| "option_a": "XGBoost gradient-boosted decision trees", |
| "option_b": "fully connected deep neural network", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0091", |
| "paper_id": "2024AJ....168..234L", |
| "question": "Among tree-based classifiers for identifying true stellar flares from flare-candidate profile features, which method better suppresses false positives without losing flare recovery?", |
| "option_a": "XGBoost gradient-boosted decision trees", |
| "option_b": "Random Forest", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0092", |
| "paper_id": "2024A&A...690A.326L", |
| "question": "For identifying galaxy mergers in high-resolution galaxy images, which CNN-based consensus classifier gives a better purity-completeness balance on visually classified real-galaxy test sets?", |
| "option_a": "dual-CNN consensus classifier with a single decision cutoff", |
| "option_b": "dual-CNN consensus classifier with ambiguous-case rejection", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0093", |
| "paper_id": "2024MNRAS.532.1902R", |
| "question": "For mitigating imaging systematics in photometric galaxy-clustering measurements used to constrain local primordial non-Gaussianity, which data-driven regression treatment better passes residual-systematics tests?", |
| "option_a": "linear multivariate regression", |
| "option_b": "feedforward neural-network regression", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0094", |
| "paper_id": "2024RAA....24d5025L", |
| "question": "For identifying cloudy-contaminated frames in light-curve preprocessing, when a classifier is trained on cloudy-frame data from one observing night and evaluated on frames from another observing night, which supervised classifier achieves higher classification accuracy?", |
| "option_a": "support-vector machine", |
| "option_b": "LightGBM", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0095", |
| "paper_id": "2024ApJ...967..152A", |
| "question": "For predicting galaxy star formation rates from broad-band galaxy images, which ResNet50-based architecture achieves higher test-set prediction accuracy?", |
| "option_a": "sequential ResNet50 CNN chain", |
| "option_b": "standalone ResNet50 CNN", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0096", |
| "paper_id": "2024MNRAS.527.1240H", |
| "question": "For inferring plane-of-sky magnetic-field orientation from spectroscopic velocity-channel maps under comparable molecular-cloud turbulence conditions, which CNN architecture gives lower median angular uncertainty?", |
| "option_a": "2D convolution-pooling CNN", |
| "option_b": "3D residual U-Net autoencoder", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0097", |
| "paper_id": "2024ApJ...964..130J", |
| "question": "For photometric redshift estimation in cosmology, which machine-learning pipeline gives lower RMS error and fewer outlier predictions?", |
| "option_a": "Random Forest", |
| "option_b": "Bayesian neural network with uncertainty-based rejection", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0098", |
| "paper_id": "2024ApJ...964..130J", |
| "question": "For photometric redshift estimation in cosmology, which machine-learning pipeline gives lower RMS error and fewer outlier predictions?", |
| "option_a": "standard neural network", |
| "option_b": "Bayesian neural network with uncertainty-based rejection", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0099", |
| "paper_id": "2024ApJ...964..130J", |
| "question": "For photometric redshift estimation in cosmology, which machine-learning pipeline gives lower RMS error and fewer outlier predictions?", |
| "option_a": "XGBoost", |
| "option_b": "Bayesian neural network with uncertainty-based rejection", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0100", |
| "paper_id": "2024A&A...686A.144B", |
| "question": "For filtering radio single-pulse candidates in magnetar observations before applying source-specific RFI rejection, which neural-network classification strategy gives the higher F-score?", |
| "option_a": "single neural-network classifier", |
| "option_b": "three-network voting ensemble", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0101", |
| "paper_id": "2024MNRAS.527.3381D", |
| "question": "For predicting baryonic galaxy properties from dark-matter halo properties in cosmological simulations, which neural-network approach gives better predictive accuracy while more faithfully reproducing the target distribution?", |
| "option_a": "physics-informed neural network with KLD distribution loss", |
| "option_b": "MSE-only multilayer perceptron", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0102", |
| "paper_id": "2024MNRAS.527.3381D", |
| "question": "For physics-informed baryon inpainting from dark-matter halo properties, which loss design better improves prediction accuracy and scatter recovery?", |
| "option_a": "SHMR-constrained PINN", |
| "option_b": "PINN with KLD distribution loss", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0103", |
| "paper_id": "2024A&A...690A.236D", |
| "question": "For emulating halo catalogues from gridded dark-matter density fields, which neural halo-bias emulator better matches N-body halo clustering statistics?", |
| "option_a": "local-density mixture-density network", |
| "option_b": "non-local symmetric-convolution mixture-density network", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0104", |
| "paper_id": "2024A&A...681A...3G", |
| "question": "For parameterizing physically consistent exoplanet atmospheric pressure-temperature profiles, which data-driven representation gives lower test-set reconstruction error?", |
| "option_a": "PCA linear-basis parameterization", |
| "option_b": "neural latent-variable decoder", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0105", |
| "paper_id": "2024A&A...683A..90Z", |
| "question": "For automatic detection and classification of type II and III solar radio bursts in dynamic spectra, which object-detection network more reliably localizes burst events while maintaining precision and recall?", |
| "option_a": "YOLOv7 object detector", |
| "option_b": "Faster R-CNN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0106", |
| "paper_id": "2024A&A...681A..86M", |
| "question": "For the multi-output observables step in stellar-evolution track emulation, which supervised regressor more accurately predicts luminosity, effective temperature, and surface gravity?", |
| "option_a": "random forest regressor", |
| "option_b": "feedforward neural network regressor", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0107", |
| "paper_id": "2024A&A...681A..86M", |
| "question": "For the multi-output observables step in stellar-evolution track emulation, which supervised regressor more accurately predicts luminosity, effective temperature, and surface gravity?", |
| "option_a": "feedforward neural network regressor", |
| "option_b": "k-nearest-neighbor regressor", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0108", |
| "paper_id": "2024ApJ...975..297W", |
| "question": "When combining mock strong-lens images to constrain the number density of low-mass dark-matter subhalos, which approach gives tighter constraints with comparable computing resources?", |
| "option_a": "sequential neural posterior estimation", |
| "option_b": "non-sequential neural posterior estimation", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0109", |
| "paper_id": "2024A&A...691A.106M", |
| "question": "Which ML approach more accurately classifies variable stars from their light curves?", |
| "option_a": "CNN + Random Forest two-stage classifier", |
| "option_b": "CNN classifier", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0110", |
| "paper_id": "2024AJ....168..191K", |
| "question": "For identifying ring galaxies in mock galaxy images, which CNN architecture better balances completeness and purity?", |
| "option_a": "ResNet-101", |
| "option_b": "Inception-ResNet-V2", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0111", |
| "paper_id": "2024AJ....168..191K", |
| "question": "When adapting a simulation-pretrained ring-galaxy classifier to a limited sample of real images, which fine-tuning strategy gives more accurate classifications?", |
| "option_a": "output-layer-only fine-tuning", |
| "option_b": "upper-block and output-layer fine-tuning", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0112", |
| "paper_id": "2024AJ....168..191K", |
| "question": "With only a limited labeled sample of real galaxy images, which CNN training strategy more accurately distinguishes ring from non-ring galaxies?", |
| "option_a": "partial fine-tuning of a simulation-pretrained CNN", |
| "option_b": "training the CNN from scratch", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0113", |
| "paper_id": "2024A&A...691A.221D", |
| "question": "For classifying survey sources as stars, galaxies, or quasars, which supervised ML approach gives more accurate classifications?", |
| "option_a": "random forest", |
| "option_b": "dropout Bayesian neural network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0114", |
| "paper_id": "2024A&A...691A.221D", |
| "question": "For classifying survey sources as stars, galaxies, or quasars, which supervised ML approach gives more accurate classifications?", |
| "option_a": "dropout Bayesian neural network", |
| "option_b": "XGBoost gradient-boosted trees", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0115", |
| "paper_id": "2024A&A...686A.209B", |
| "question": "For repeated predictions of the linear matter power spectrum from cosmological parameters, which ML emulator is faster to evaluate?", |
| "option_a": "neural-network emulator", |
| "option_b": "genetic-programming symbolic-regression emulator", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0116", |
| "paper_id": "2024ApJS..272....4Z", |
| "question": "For identifying gamma-ray bursts in gamma-ray detector count maps, which CNN architecture gives more accurate predictions?", |
| "option_a": "plain CNN", |
| "option_b": "residual CNN with spatial-and-channel attention", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0117", |
| "paper_id": "2024A&A...683A.185T", |
| "question": "Given the parameters of a synchrotron self-Compton emission model, which machine-learning method more accurately predicts a blazar's spectral energy distribution?", |
| "option_a": "gated recurrent unit network", |
| "option_b": "feed-forward neural network", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0118", |
| "paper_id": "2024A&A...683A.185T", |
| "question": "Given the parameters of a synchrotron self-Compton emission model, which machine-learning method more accurately predicts a blazar's spectral energy distribution?", |
| "option_a": "feed-forward neural network", |
| "option_b": "long short-term memory network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0119", |
| "paper_id": "2024A&A...683A.185T", |
| "question": "Given the parameters of a synchrotron self-Compton emission model, which machine-learning method better preserves the overall shape of the predicted blazar spectral energy distribution?", |
| "option_a": "long short-term memory network", |
| "option_b": "gated recurrent unit network", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0120", |
| "paper_id": "2024MNRAS.533.2547F", |
| "question": "For identifying galaxy mergers near coalescence in realistic mock optical images, which ML approach recovers more true mergers while maintaining comparable sample purity?", |
| "option_a": "hybrid CNN-Vision Transformer ensemble", |
| "option_b": "single CNN classifier", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0121", |
| "paper_id": "2024MNRAS.533.2547F", |
| "question": "For identifying galaxy mergers in realistic mock optical images, which ML approach produces a less contaminated candidate sample?", |
| "option_a": "majority-vote ensemble of CNN and Vision Transformer models", |
| "option_b": "full-consensus ensemble of CNN and Vision Transformer models", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0122", |
| "paper_id": "2024MNRAS.531.4203B", |
| "question": "For a fixed photometric-survey configuration, which ML approach predicts tomographic angular two-point statistics more quickly?", |
| "option_a": "matter-power-spectrum neural-network emulator + numerical angular projection", |
| "option_b": "direct neural-network emulator of angular statistics", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0123", |
| "paper_id": "2024MNRAS.527..651T", |
| "question": "For estimating galaxy redshifts from broadband photometric observations, which machine-learning pipeline yields lower robust scatter and fewer catastrophic estimates?", |
| "option_a": "magnitude-and-color k-nearest-neighbor regression", |
| "option_b": "image-based convolutional neural network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0124", |
| "paper_id": "2024MNRAS.527..651T", |
| "question": "For estimating galaxy redshifts from multiband images when spectroscopic redshifts are available for relatively few high-redshift training galaxies, which CNN training strategy yields lower prediction scatter?", |
| "option_a": "redshift regression alone", |
| "option_b": "joint redshift-bin classification and regression", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0125", |
| "paper_id": "2024A&A...683A..26A", |
| "question": "For estimating galaxy redshifts from multiband images, which CNN architecture yields more precise estimates?", |
| "option_a": "multi-stream CNN", |
| "option_b": "single-stream CNN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0126", |
| "paper_id": "2024A&A...683A..26A", |
| "question": "For estimating galaxy redshifts from optical and near-infrared images, which CNN approach yields more precise estimates?", |
| "option_a": "CNN classification over redshift intervals", |
| "option_b": "direct CNN redshift regression", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0127", |
| "paper_id": "2024MNRAS.527..739R", |
| "question": "For estimating warm-dark-matter particle masses from cosmological simulations, which neural-network pipeline gives lower prediction error?", |
| "option_a": "CNN using matter-density images", |
| "option_b": "fully connected neural network using matter power spectra", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0128", |
| "paper_id": "2024ApJ...965..104F", |
| "question": "Which machine-learning method more accurately reconstructs simulated multiband quasar light curves with sparse sampling?", |
| "option_a": "multioutput Gaussian process regression", |
| "option_b": "neural stochastic differential-equation model", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0129", |
| "paper_id": "2024ApJ...965..101C", |
| "question": "Which machine-learning pipeline more accurately predicts galaxy stellar mass, gas mass, and stellar metallicity from dark-matter halo properties?", |
| "option_a": "multilayer perceptron using individual halo properties", |
| "option_b": "graph neural network using complete merger trees", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0130", |
| "paper_id": "2024A&A...687A..45P", |
| "question": "For estimating the time before or after a galaxy merger from simulated multiband images, which neural-network architecture gives lower prediction error?", |
| "option_a": "50-layer residual CNN", |
| "option_b": "six-layer CNN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0131", |
| "paper_id": "2024A&A...687A..45P", |
| "question": "For estimating the time before or after a galaxy merger from simulated multiband images, which complete ML pipeline gives lower overall prediction error?", |
| "option_a": "u, g, r, and i-band CNN regression", |
| "option_b": "u, g, and r-band ImageNet-pretrained Swin Transformer regression", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0132", |
| "paper_id": "2024A&A...683A.145F", |
| "question": "For identifying the edges of galaxies in deep images, which ML approach more closely matches astronomers' measurements?", |
| "option_a": "majority-vote U-Net ensemble", |
| "option_b": "single U-Net", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0133", |
| "paper_id": "2024AJ....167..264L", |
| "question": "For recovering strong-lens candidates from noisy multiband images, which ML pipeline reaches higher completeness?", |
| "option_a": "Swin Transformer detector without neural image restoration", |
| "option_b": "neural image-restoration model + Swin Transformer detector", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0134", |
| "paper_id": "2024ApJS..271...54F", |
| "question": "For estimating quasar photometric redshifts, which ML approach gives more precise estimates with fewer catastrophic failures?", |
| "option_a": "XGBoost regression", |
| "option_b": "XGBoost/TabNet/FT-Transformer ensemble", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0135", |
| "paper_id": "2024MNRAS.530.1129P", |
| "question": "For assigning binary-black-hole mergers to galaxies in cosmological simulations, which neural-network formulation more faithfully reproduces the merger-count distribution across galaxies?", |
| "option_a": "likelihood-trained distribution-output neural network", |
| "option_b": "MSE-trained point-output neural network", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0136", |
| "paper_id": "2024MNRAS.531.1534C", |
| "question": "Which ML pipeline more accurately distinguishes clustering dark energy from a cosmological constant in simulated matter structure?", |
| "option_a": "3D density-field CNN pipeline", |
| "option_b": "power-spectrum random-forest pipeline", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0137", |
| "paper_id": "2024MNRAS.531.1534C", |
| "question": "For clustering dark-energy models away from the cosmological-constant limit, which ML pipeline more accurately distinguishes different sound speeds when matter clustering is sampled over large scales?", |
| "option_a": "3D density-field CNN pipeline", |
| "option_b": "power-spectrum random-forest pipeline", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0138", |
| "paper_id": "2024ApJ...976..200M", |
| "question": "For inferring disk and planet properties from simulated two-dimensional disk maps, which ML pipeline recovers the parameters more accurately?", |
| "option_a": "direct CNN parameter regression", |
| "option_b": "neural-operator forward modeling with evolutionary parameter search", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0139", |
| "paper_id": "2024A&A...691A.218S", |
| "question": "For estimating effective temperature from low-resolution spectra of late-F to early-K stars, which ML method gives lower mean-squared prediction error?", |
| "option_a": "multilayer perceptron", |
| "option_b": "one-dimensional convolutional neural network", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0140", |
| "paper_id": "2024A&A...691A.218S", |
| "question": "For estimating effective temperature from low-resolution spectra of late-F to early-K stars, which ML method gives lower mean-squared prediction error?", |
| "option_a": "one-dimensional convolutional neural network", |
| "option_b": "Random Forest regression", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0141", |
| "paper_id": "2024A&A...691A.218S", |
| "question": "For estimating effective temperature from low-resolution spectra of late-F to early-K stars, which ML method gives lower mean-squared prediction error?", |
| "option_a": "one-dimensional convolutional neural network", |
| "option_b": "k-nearest-neighbor regression", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0142", |
| "paper_id": "2024ApJ...977..223L", |
| "question": "For estimating black hole mass, accretion rate, and accretion-disk inclination from simulated multiband AGN light curves, which neural-network architecture yields smaller parameter uncertainties?", |
| "option_a": "CNN", |
| "option_b": "LSTM", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0143", |
| "paper_id": "2024ApJ...977..223L", |
| "question": "For estimating black hole mass, accretion rate, and accretion-disk inclination from simulated multiband AGN light curves, which neural inference pipeline yields smaller parameter uncertainties?", |
| "option_a": "LSTM followed by a normalizing-flow estimator", |
| "option_b": "normalizing-flow estimator applied directly to the light curves", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0144", |
| "paper_id": "2024A&A...684A.203B", |
| "question": "For emulating the time evolution of non-equilibrium interstellar-medium chemistry across varying initial conditions, which neural method gives more accurate gas-temperature and chemical-abundance predictions?", |
| "option_a": "physics-informed neural network", |
| "option_b": "branch-trunk neural operator", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0145", |
| "paper_id": "2024A&A...689A.274E", |
| "question": "For predicting detailed galaxy morphology from galaxy images, which CNN training strategy more accurately reproduces visual morphology classifications?", |
| "option_a": "transfer-learned CNN", |
| "option_b": "CNN trained from scratch", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0146", |
| "paper_id": "2024MNRAS.530.2012T", |
| "question": "For estimating galaxy photometric redshifts from multi-band fluxes and uncertainties, which neural-network architecture gives more precise estimates?", |
| "option_a": "one-dimensional convolutional neural network", |
| "option_b": "fully connected neural network", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0147", |
| "paper_id": "2024MNRAS.530.2012T", |
| "question": "When only a small spectroscopic training sample is available, which CNN training strategy gives more precise photometric-redshift estimates?", |
| "option_a": "CNN trained from scratch on the spectroscopic sample", |
| "option_b": "simulation-pretrained CNN with spectroscopic fine-tuning", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0148", |
| "paper_id": "2024MNRAS.530.2012T", |
| "question": "For obtaining well-calibrated photometric-redshift probability distributions, which neural-network approach performs better?", |
| "option_a": "single convolutional neural network", |
| "option_b": "ensemble of independently trained convolutional neural networks", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0149", |
| "paper_id": "2024ApJS..274...29C", |
| "question": "For distinguishing variable-star classes and transient types from their light curves, which CNN pipeline gives more accurate classifications?", |
| "option_a": "CNN combining raw light curves with periodogram-guided phase-folded light curves", |
| "option_b": "CNN using periodogram-guided phase-folded light curves alone", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0150", |
| "paper_id": "2024ApJS..274...29C", |
| "question": "For distinguishing variable-star classes from their light curves, which CNN pipeline gives more accurate classifications?", |
| "option_a": "CNN with learned frequency-window selection and phase folding", |
| "option_b": "CNN with learned time-window selection", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0151", |
| "paper_id": "2024ApJS..274...29C", |
| "question": "For distinguishing variable-star classes directly from their light curves, which CNN architecture is more accurate?", |
| "option_a": "CNN without learned time-window selection", |
| "option_b": "CNN with learned time-window selection", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0152", |
| "paper_id": "2024MNRAS.530...20A", |
| "question": "For identifying galaxy-cluster mergers from idealized Compton-y and X-ray maps, which CNN pipeline detects a larger fraction of true mergers?", |
| "option_a": "single-branch CNN using Compton-y maps", |
| "option_b": "two-branch CNN combining Compton-y and X-ray maps", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0153", |
| "paper_id": "2024MNRAS.530...20A", |
| "question": "For identifying galaxy-cluster mergers from idealized Compton-y and X-ray maps, which CNN pipeline detects a larger fraction of true mergers?", |
| "option_a": "two-branch CNN combining Compton-y and X-ray maps", |
| "option_b": "single-branch CNN using X-ray maps", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0154", |
| "paper_id": "2024ApJ...975...38L", |
| "question": "How can cosmological parameters be more precisely constrained from the sky positions, redshifts, and stellar masses of galaxies in an observed redshift survey?", |
| "option_a": "cross-domain representation-alignment PointNet", |
| "option_b": "standard supervised PointNet", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0155", |
| "paper_id": "2024MNRAS.527.2585L", |
| "question": "How can a Be star's inclination angle be accurately inferred from a single observed Halpha emission-line profile?", |
| "option_a": "feed-forward neural-network regression", |
| "option_b": "feed-forward neural-network classification over inclination bins", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0156", |
| "paper_id": "2024AJ....168...38H", |
| "question": "For modeling the completeness of two-band stellar photometry from artificial-star tests, which machine-learning method is more accurate?", |
| "option_a": "multilayer perceptron", |
| "option_b": "logistic regression", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0157", |
| "paper_id": "2024A&A...683A.105D", |
| "question": "For detecting faint point sources in simulated radio-interferometric observations, which ML pipeline achieves higher completeness?", |
| "option_a": "dirty-image conditional denoising diffusion pipeline", |
| "option_b": "visibility-domain fully connected network with autoencoder refinement", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0158", |
| "paper_id": "2024MNRAS.534..385W", |
| "question": "For detecting tracked targets in non-sidereal astronomical image sequences, which ML pipeline recovers more true targets with fewer false detections?", |
| "option_a": "image-registration and K-means classification with track association", |
| "option_b": "fully connected neural-network classification with track association", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0159", |
| "paper_id": "2024MNRAS.529.1654T", |
| "question": "For inferring accretion-flow parameters from black-hole X-ray binary spectra, which neural-network training strategy produces more accurate spectral fits?", |
| "option_a": "joint parameter-and-reconstruction learning", |
| "option_b": "parameter-only supervised learning", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0160", |
| "paper_id": "2024MNRAS.529.1654T", |
| "question": "When training on simulated black-hole X-ray binary spectra, which learning strategy produces more accurate spectral fits?", |
| "option_a": "reconstruction-only learning", |
| "option_b": "parameter-only supervised learning", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0161", |
| "paper_id": "2024MNRAS.529.1654T", |
| "question": "When training on simulated black-hole X-ray binary spectra, which learning strategy produces more accurate spectral fits?", |
| "option_a": "reconstruction-only learning", |
| "option_b": "joint parameter-and-reconstruction learning", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0162", |
| "paper_id": "2024A&A...682A...4T", |
| "question": "For separating low-surface-brightness galaxies from image artifacts, which Transformer architecture yields lower contamination while retaining similar completeness?", |
| "option_a": "self-attention-only vision Transformer ensemble", |
| "option_b": "convolution-augmented Transformer ensemble", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0163", |
| "paper_id": "2024AJ....167...73S", |
| "question": "For estimating stellar distances from spectroscopic measurements, which neural-network pipeline gives more consistent distances for stars in the same cluster?", |
| "option_a": "compact dense network using derived stellar parameters and abundances", |
| "option_b": "Bayesian convolutional network using full spectra", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0164", |
| "paper_id": "2024MNRAS.52711559N", |
| "question": "For estimating missing stellar line-of-sight velocities from astrometric measurements, which neural-network approach yields narrower prediction intervals with a lower calibration error?", |
| "option_a": "deterministic feed-forward neural network", |
| "option_b": "Bayesian neural network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0165", |
| "paper_id": "2024MNRAS.528...82K", |
| "question": "For classifying spiral and elliptical galaxies when the training labels are generated by unsupervised clustering, which ML pipeline gives higher and more balanced accuracy on independent samples?", |
| "option_a": "agglomerative-clustering labeling pipeline + CNN", |
| "option_b": "self-organizing-map labeling pipeline + CNN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0166", |
| "paper_id": "2024MNRAS.528...82K", |
| "question": "For classifying spiral and elliptical galaxies when the training labels are generated by unsupervised clustering, which ML pipeline gives higher and more balanced accuracy on independent samples?", |
| "option_a": "self-organizing-map labeling pipeline + CNN", |
| "option_b": "spectral-clustering labeling pipeline + CNN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0167", |
| "paper_id": "2024AJ....167...29W", |
| "question": "For classifying galaxy morphology from optical images, which CNN architecture gives higher accuracy?", |
| "option_a": "deformable-convolution residual CNN with layer attention", |
| "option_b": "standard residual CNN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0168", |
| "paper_id": "2024AJ....167..260W", |
| "question": "For classifying spectra of stars, galaxies, and quasars, which CNN architecture gives higher accuracy?", |
| "option_a": "ConvNeXt CNN", |
| "option_b": "multiscale partial-convolution CNN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0169", |
| "paper_id": "2024AJ....167..260W", |
| "question": "For classifying stellar subclasses from spectra, which convolutional design gives higher accuracy?", |
| "option_a": "multiscale large-kernel convolution", |
| "option_b": "small-kernel convolution", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0170", |
| "paper_id": "2024ApJ...967..155L", |
| "question": "For identifying pulsar candidates with limited labeled data, which pseudolabeling strategy gives a higher F1 score?", |
| "option_a": "confidence-weighted pseudolabeling with a validation-adjusted threshold", |
| "option_b": "fixed-threshold pseudolabeling", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0171", |
| "paper_id": "2024ApJ...967..155L", |
| "question": "For identifying pulsar candidates with limited labeled data, which semisupervised pipeline gives a higher F1 score?", |
| "option_a": "semi-supervised generative adversarial network", |
| "option_b": "confidence-weighted pseudolabeling WideResNet", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0172", |
| "paper_id": "2024ApJ...970..168W", |
| "question": "For estimating photospheric magnetic flux density from Stokes observations of strong-field regions, which neural-network training strategy gives lower prediction error?", |
| "option_a": "mean-squared-error regression", |
| "option_b": "classification-based regression", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0173", |
| "paper_id": "2024ApJS..271....9D", |
| "question": "For increasing the spatial resolution of solar magnetograms, which deep-learning architecture better preserves magnetic-field structure?", |
| "option_a": "multibranch residual-dense CNN", |
| "option_b": "generative adversarial network", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0174", |
| "paper_id": "2024ApJS..271....9D", |
| "question": "For increasing the spatial resolution of solar magnetograms at similar reconstruction accuracy, which CNN architecture requires less computation?", |
| "option_a": "fixed-complexity residual-dense CNN", |
| "option_b": "region-routed multibranch residual-dense CNN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0175", |
| "paper_id": "2024ApJS..271...31Z", |
| "question": "For predicting when a coronal mass ejection will reach Earth from coronagraph images, which CNN architecture gives lower prediction error?", |
| "option_a": "VGG CNN", |
| "option_b": "ResNet-18", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0176", |
| "paper_id": "2024ApJS..271...31Z", |
| "question": "For predicting when a coronal mass ejection will reach Earth, which ML pipeline gives lower prediction error?", |
| "option_a": "attention-fusion neural network using physical parameters and coronagraph images", |
| "option_b": "support-vector regression using physical parameters", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0177", |
| "paper_id": "2024ApJS..271...46M", |
| "question": "For increasing the spatial resolution of solar magnetograms while preserving strong magnetic fields and magnetic-flux balance, which CNN training objective is more effective?", |
| "option_a": "mean-squared-error loss with gradient, field-distribution, and structural-similarity penalties", |
| "option_b": "mean-squared-error loss alone", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0178", |
| "paper_id": "2024MNRAS.527.1163W", |
| "question": "For classifying galaxies by emission-line activity from their spectra, which ML method gives higher accuracy?", |
| "option_a": "one-dimensional convolutional neural network", |
| "option_b": "random forest", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0179", |
| "paper_id": "2024MNRAS.527.9930Z", |
| "question": "For identifying stars with Jupiter-like planets from high-resolution spectra, which ML pipeline gives more accurate classifications?", |
| "option_a": "stacked ensemble of convolutional neural networks", |
| "option_b": "stacked random-forest ensemble", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0180", |
| "paper_id": "2024MNRAS.528..873S", |
| "question": "For detecting low-surface-brightness galaxies in images, which object-detection architecture gives a better balance of completeness and purity?", |
| "option_a": "coordinate-attention CNN detector", |
| "option_b": "YOLOv4 CNN detector", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0181", |
| "paper_id": "2024MNRAS.528..873S", |
| "question": "For detecting low-surface-brightness galaxies with limited labeled images, which CNN training strategy gives higher completeness and purity?", |
| "option_a": "training the same CNN from random initialization", |
| "option_b": "transfer learning from a pretrained object-detection CNN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0182", |
| "paper_id": "2024RAA....24d5020P", |
| "question": "For classifying galaxies as spiral or elliptical from infrared data, which ML pipeline gives higher accuracy?", |
| "option_a": "support-vector machine using infrared colors", |
| "option_b": "CNN using infrared galaxy images", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0183", |
| "paper_id": "2024RAA....24d5020P", |
| "question": "For classifying galaxies as spiral or elliptical from infrared images, which CNN architecture gives higher accuracy?", |
| "option_a": "residual CNN", |
| "option_b": "CNN with channel and spatial attention", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0184", |
| "paper_id": "2024RAA....24d5020P", |
| "question": "For classifying galaxies as spiral or elliptical, which CNN pipeline gives higher accuracy?", |
| "option_a": "stacked CNN combining infrared and optical images", |
| "option_b": "CNN using optical images", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0185", |
| "paper_id": "2024RAA....24e5013L", |
| "question": "For detecting anomalous variations in stellar light curves, which GRU architecture gives a higher F1 score?", |
| "option_a": "bidirectional GRU with attention", |
| "option_b": "GRU without attention", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0186", |
| "paper_id": "2024RAA....24f5024B", |
| "question": "For estimating active-surface actuator adjustments from simulated radio-telescope reflector deformations, which CNN design gives lower prediction error?", |
| "option_a": "CNN with learnable panel-node weighting", |
| "option_b": "CNN without learnable panel-node weighting", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0187", |
| "paper_id": "2024RAA....24i5015S", |
| "question": "For increasing the resolution of simulated telescope images, which ML pipeline gives more accurate reconstructions?", |
| "option_a": "cascaded-residual GAN trained with a fixed telescope PSF", |
| "option_b": "cascaded-residual GAN trained with a learnable telescope PSF", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0188", |
| "paper_id": "2024RAA....24j5010Z", |
| "question": "For extrapolating force-free coronal magnetic fields, which neural-network architecture converges more reliably?", |
| "option_a": "physics-informed neural network with a two-stage output layer", |
| "option_b": "physics-informed neural network with a single output layer", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0189", |
| "paper_id": "2025ApJ...994...62C", |
| "question": "For estimating warm dark matter particle mass from simulated galaxy populations, which ML method gives more accurate estimates?", |
| "option_a": "permutation-invariant set network + normalizing flow", |
| "option_b": "graph neural network + normalizing flow", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0190", |
| "paper_id": "2025ApJ...994...62C", |
| "question": "For estimating warm dark matter particle mass from simulated galaxy populations, which ML pipeline gives lower prediction error?", |
| "option_a": "graph neural network using individual subhalo properties + normalizing flow", |
| "option_b": "multilayer perceptron using population-level total-mass statistics + normalizing flow", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0191", |
| "paper_id": "2025ApJ...981...52C", |
| "question": "For reconstructing mass distributions from simulated wide-field weak-lensing shear fields, which CNN pipeline produces more faithful convergence maps?", |
| "option_a": "parallel-multikernel CNN pipeline", |
| "option_b": "fixed-large-kernel CNN pipeline", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0192", |
| "paper_id": "2025ApJ...981...52C", |
| "question": "For detecting low-mass galaxy clusters from simulated weak-lensing observations, which CNN pipeline achieves higher completeness?", |
| "option_a": "parallel-multikernel CNN pipeline", |
| "option_b": "fixed-large-kernel CNN pipeline", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0193", |
| "paper_id": "2025A&A...696A..77T", |
| "question": "For estimating spectral-model parameters from simulated X-ray spectra, which method gives lower prediction errors?", |
| "option_a": "gated recurrent neural network", |
| "option_b": "multilayer perceptron", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0194", |
| "paper_id": "2025A&A...696A..77T", |
| "question": "For estimating spectral-model parameters from simulated X-ray spectra, which method gives lower prediction errors?", |
| "option_a": "hybrid convolutional and dense neural network", |
| "option_b": "gated recurrent neural network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0195", |
| "paper_id": "2025AJ....169..201T", |
| "question": "For classifying asteroids from visible and near-infrared reflectance spectra, which ML method gives higher overall classification accuracy?", |
| "option_a": "convolutional network with multi-head self-attention", |
| "option_b": "Random Forest", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0196", |
| "paper_id": "2025AJ....169..201T", |
| "question": "For classifying asteroids from visible and near-infrared reflectance spectra, which neural-network architecture gives higher overall classification accuracy?", |
| "option_a": "ResNet-50 CNN", |
| "option_b": "convolutional network with multi-head self-attention", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0197", |
| "paper_id": "2025AJ....169..201T", |
| "question": "For estimating mineral spectra and mineral proportions from simulated S-type asteroid spectra, which neural-network architecture gives lower reconstruction errors?", |
| "option_a": "Hapke-informed convolutional encoder-decoder", |
| "option_b": "convolutional autoencoder with Transformer attention", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0198", |
| "paper_id": "2025A&A...702A.245T", |
| "question": "For recovering turnaround radii from the collective galaxy kinematics of simulated cluster samples grouped by similar dynamical scales, which machine-learning pipeline gives lower prediction error?", |
| "option_a": "CNN on two-dimensional velocity-dispersion maps", |
| "option_b": "XGBoost on radial velocity-dispersion profiles", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0199", |
| "paper_id": "2025MNRAS.544.2586N", |
| "question": "For reconstructing small-scale features of the early cosmic matter distribution from the present-day density field in simulations, which 3D CNN architecture is more accurate?", |
| "option_a": "single-scale 3D CNN", |
| "option_b": "dual-scale 3D CNN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0200", |
| "paper_id": "2025A&A...694A.161S", |
| "question": "For estimating a star's effective temperature from a single flare light curve, which regression method is more accurate?", |
| "option_a": "linear regression", |
| "option_b": "gradient-boosted decision trees", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0201", |
| "paper_id": "2025RAA....25d4008S", |
| "question": "For distinguishing transients from image-subtraction artifacts, which CNN approach gives more reliable classifications?", |
| "option_a": "DenseNet CNN", |
| "option_b": "ensemble of ResNet, DenseNet, and EfficientNet CNNs", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0202", |
| "paper_id": "2025RAA....25d4008S", |
| "question": "For distinguishing transients from image-subtraction artifacts, which CNN architecture produces a less contaminated candidate sample?", |
| "option_a": "EfficientNet CNN", |
| "option_b": "ResNet CNN", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0203", |
| "paper_id": "2025RAA....25d4008S", |
| "question": "For distinguishing transients from image-subtraction artifacts, which CNN architecture achieves higher transient completeness?", |
| "option_a": "ResNet CNN", |
| "option_b": "EfficientNet CNN", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0204", |
| "paper_id": "2025ApJS..279...27H", |
| "question": "For forecasting large solar flares from a uniformly sampled X-ray flare time series, which ML approach better separates large from smaller events?", |
| "option_a": "bagging ensemble of LSTMs with moving-average decomposition", |
| "option_b": "single LSTM with moving-average decomposition", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0205", |
| "paper_id": "2025ApJS..278....4W", |
| "question": "For locating and classifying solar radio bursts in radio spectrograms, which attention mechanism gives more accurate detections?", |
| "option_a": "deformable attention with a dynamic-convolution adapter", |
| "option_b": "standard deformable attention", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0206", |
| "paper_id": "2025ApJS..278....4W", |
| "question": "For locating and classifying solar radio bursts in radio spectrograms, which Transformer detector variant achieves higher recall?", |
| "option_a": "Transformer detector trained with one-to-many auxiliary matching", |
| "option_b": "Transformer detector trained with one-to-one matching", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0207", |
| "paper_id": "2025ApJS..278....4W", |
| "question": "For locating and classifying solar radio bursts in radio spectrograms, which object-detection pipeline gives more accurate detections while missing fewer bursts?", |
| "option_a": "dynamic-convolution Transformer detector with auxiliary matching", |
| "option_b": "convolution-Transformer single-shot detector", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0208", |
| "paper_id": "2025MNRAS.541.3269M", |
| "question": "For estimating weak local-type primordial non-Gaussianity from higher-resolution simulated full-sky cosmic microwave background maps, which method gives lower prediction error?", |
| "option_a": "graph-convolution CNN", |
| "option_b": "pixel-neighborhood convolution CNN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0209", |
| "paper_id": "2025ApJ...995..136Z", |
| "question": "For estimating the stellar parameters of O stars from their light curves, which method is more accurate?", |
| "option_a": "one-dimensional convolutional neural network", |
| "option_b": "multilayer perceptron", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0210", |
| "paper_id": "2025MNRAS.542..426T", |
| "question": "For reconstructing simulated wide-field radio-interferometric observations, which ML pipeline produces more accurate images?", |
| "option_a": "residual U-Net series with planar-domain loss and final spherical projection", |
| "option_b": "residual U-Net series with spherical-domain loss and iterative spherical projection", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0211", |
| "paper_id": "2025A&A...697A.162L", |
| "question": "When inferring cosmological parameters from simulated weak-lensing maps, which CNN training strategy yields tighter constraints?", |
| "option_a": "CNN trained by mutual-information maximization", |
| "option_b": "CNN trained by mean-squared-error regression", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0212", |
| "paper_id": "2025A&A...697A.162L", |
| "question": "When constraining dark energy with simulated weak-lensing maps, which CNN training strategy yields tighter constraints?", |
| "option_a": "CNN trained by mean-absolute-error regression", |
| "option_b": "CNN trained by mutual-information maximization", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0213", |
| "paper_id": "2025A&A...697A.162L", |
| "question": "When inferring matter density and clustering amplitude from simulated weak-lensing maps, which CNN training strategy yields tighter constraints?", |
| "option_a": "CNN trained by mutual-information maximization", |
| "option_b": "CNN trained to fit a Gaussian posterior", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0214", |
| "paper_id": "2025ApJ...985L...7J", |
| "question": "For recovering solar magnetic fields from noisy simulated spectropolarimetric time series, which neural-network pipeline gives more accurate inversions?", |
| "option_a": "joint spatiotemporal physics-informed neural network", |
| "option_b": "frame-by-frame physics-informed neural network", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0215", |
| "paper_id": "2025MNRAS.538..133P", |
| "question": "For rejecting image-subtraction artifacts in transient searches, which CNN pipeline yields a cleaner candidate sample while retaining most transients?", |
| "option_a": "single-stage CNN artifact classifier", |
| "option_b": "source-morphology-routed CNN pipeline", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0216", |
| "paper_id": "2025MNRAS.537.3383T", |
| "question": "For modeling the velocity evolution of infalling cold clouds in hydrodynamic simulations, which neural-ODE design gives lower prediction error?", |
| "option_a": "neural ODE with a single network for cloud growth", |
| "option_b": "neural ODE with separate networks for cloud growth and drag", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0217", |
| "paper_id": "2025MNRAS.537.3383T", |
| "question": "For modeling the mass growth of infalling cold clouds in hydrodynamic simulations, which ML approach gives lower prediction error?", |
| "option_a": "symbolic-regression approximation of the neural ODE", |
| "option_b": "neural ODE", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0218", |
| "paper_id": "2025AJ....170...44E", |
| "question": "For estimating the mean lens-galaxy mass-profile slope from simulated images of strongly lensed quasars, which neural posterior-estimation method gives lower fractional error and tighter constraints?", |
| "option_a": "single-stage neural posterior estimation", |
| "option_b": "sequential neural posterior estimation", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0219", |
| "paper_id": "2025PhRvD.112b2003M", |
| "question": "For detecting short gravitational-wave transients in real detector data, which ML pipeline gives higher detection efficiency at a fixed false-alarm rate?", |
| "option_a": "recurrent-autoencoder detector with an auxiliary neural classifier", |
| "option_b": "recurrent-autoencoder detector alone", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0220", |
| "paper_id": "2025A&A...700A.129B", |
| "question": "For recovering galaxy flux and morphology from simulated blended images, which ML method is more accurate?", |
| "option_a": "variational autoencoder with a normalizing-flow prior and latent-space optimization", |
| "option_b": "feed-forward variational autoencoder", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0221", |
| "paper_id": "2025ApJ...991..124E", |
| "question": "For estimating redshifts of nearby galaxies from broadband observations, which ML pipeline gives lower scatter and outlier rate?", |
| "option_a": "local linear regression on catalog photometry", |
| "option_b": "CNN on galaxy image cutouts", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0222", |
| "paper_id": "2025ApJ...991..124E", |
| "question": "For estimating galaxy redshifts from broadband images, which neural-network pipeline gives lower scatter and outlier rate in its point estimates?", |
| "option_a": "image CNN followed by a monotonic neural calibration network", |
| "option_b": "image CNN", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0223", |
| "paper_id": "2025A&A...693A..73G", |
| "question": "For estimating galaxy redshifts from simulated spectra, which approach produces fewer catastrophic redshift errors?", |
| "option_a": "redshift-only CNN", |
| "option_b": "CNN with spectral-line and galaxy-property auxiliary tasks", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0224", |
| "paper_id": "2025ApJ...980..183W", |
| "question": "For estimating gas metallicity from optical galaxy images, which ML pipeline gives lower prediction error?", |
| "option_a": "CNN with PCA-compressed features", |
| "option_b": "sparsity-constrained CNN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0225", |
| "paper_id": "2025MNRAS.541.2815K", |
| "question": "For modeling the broadband flux distribution of contaminants in high-redshift quasar searches, which density-estimation method trains faster while giving comparable reconstructions?", |
| "option_a": "neural conditional Gaussian-mixture deconvolution", |
| "option_b": "binned Gaussian-mixture extreme deconvolution", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0226", |
| "paper_id": "2025MNRAS.543..691Z", |
| "question": "For estimating galaxy redshifts from spectra, which reconstruction approach achieves a higher redshift success rate?", |
| "option_a": "autoencoder-like spectral-basis network", |
| "option_b": "U-Net", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0227", |
| "paper_id": "2025A&C....5200963S", |
| "question": "For classifying galaxy morphology from images, which dimensionality-reduction pipeline gives higher accuracy?", |
| "option_a": "uniform manifold approximation and projection + logistic regression", |
| "option_b": "locally linear embedding + logistic regression", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0228", |
| "paper_id": "2025A&C....5200963S", |
| "question": "For distinguishing round, intermediate, and elongated galaxies from images, which dimensionality-reduction pipeline gives higher accuracy?", |
| "option_a": "principal component analysis + logistic regression", |
| "option_b": "locally linear embedding + logistic regression", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0229", |
| "paper_id": "2025A&C....5200963S", |
| "question": "For classifying galaxy morphology from images, which method gives higher accuracy?", |
| "option_a": "convolutional neural network", |
| "option_b": "multilayer perceptron", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0230", |
| "paper_id": "2025ApJ...980..176Z", |
| "question": "For segmenting solar filaments in full-disk hydrogen-alpha images, which U-Net architecture recovers more filament structure and agrees better with the reference masks?", |
| "option_a": "classical U-Net with expanding channel width", |
| "option_b": "channel-attention U-Net with fixed channel width", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0231", |
| "paper_id": "2025ApJ...980..176Z", |
| "question": "For segmenting solar filaments in full-disk hydrogen-alpha images, which channel-attention U-Net variant gives higher recall and better agreement with the reference masks?", |
| "option_a": "U-Net with cross-channel attention in the bottleneck", |
| "option_b": "U-Net with within-channel attention throughout", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0232", |
| "paper_id": "2025A&A...693A..42G", |
| "question": "For inferring exoplanet atmospheric properties from simulated spectra, which method trains faster while retaining comparable posterior accuracy?", |
| "option_a": "continuous normalizing flow trained by flow matching", |
| "option_b": "discrete normalizing flow trained by neural posterior estimation", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0233", |
| "paper_id": "2025A&A...693A..42G", |
| "question": "For inferring exoplanet atmospheric properties from simulated spectra, which method generates posterior samples faster?", |
| "option_a": "continuous normalizing flow trained by flow matching", |
| "option_b": "discrete normalizing flow trained by neural posterior estimation", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0234", |
| "paper_id": "2025A&A...704A.150A", |
| "question": "For identifying stars with unresolved companions from simulated astrometric fit statistics, which supervised classifier gives a better balance between completeness and purity?", |
| "option_a": "logistic regression", |
| "option_b": "multilayer perceptron", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0235", |
| "paper_id": "2025A&A...704A.150A", |
| "question": "For identifying stars with unresolved companions from simulated astrometric fit statistics, which supervised classifier gives a better balance between completeness and purity?", |
| "option_a": "multilayer perceptron", |
| "option_b": "shallow random forest", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0236", |
| "paper_id": "2025ApJ...985..132C", |
| "question": "For estimating cosmological parameters from simulated dark-matter halo catalogs with halo masses and kinematics, which ML method gives lower prediction error?", |
| "option_a": "hierarchical point-cloud MLP", |
| "option_b": "XGBoost", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0237", |
| "paper_id": "2025ApJ...985..132C", |
| "question": "For estimating the amplitude of matter fluctuations from simulated dark-matter halo positions, which ML pipeline gives lower prediction error?", |
| "option_a": "hierarchical point-cloud MLP", |
| "option_b": "two-point correlation function + multilayer perceptron", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0238", |
| "paper_id": "2025ApJ...978...64M", |
| "question": "For estimating cosmological parameters from simulated dark-matter density fields, which ML pipeline yields tighter posterior constraints?", |
| "option_a": "diffusion model + Hamiltonian Monte Carlo", |
| "option_b": "power spectrum + normalizing flow", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0239", |
| "paper_id": "2025MNRAS.536.3123J", |
| "question": "For detecting stellar flares in light curves, which neural-network pipeline performs better?", |
| "option_a": "sliding-window convolutional network", |
| "option_b": "point-wise fully convolutional network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0240", |
| "paper_id": "2025MNRAS.536.3123J", |
| "question": "For identifying flare points in stellar light curves, which loss function performs better?", |
| "option_a": "standard Dice loss", |
| "option_b": "masked Dice loss", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0241", |
| "paper_id": "2025ApJ...982...55L", |
| "question": "For predicting the distribution of gravitational-wave background spectra from simulated supermassive black-hole binary populations, which ML method is more accurate?", |
| "option_a": "conditional normalizing flow", |
| "option_b": "Gaussian process", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0242", |
| "paper_id": "2025MNRAS.536.1303J", |
| "question": "For constraining matter clustering and dark energy from weak-lensing observations, which neural-compression pipeline gives tighter posterior constraints?", |
| "option_a": "neural-network compression of power spectra alone", |
| "option_b": "convolutional neural-network compression of lensing maps combined with power-spectrum compression", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0243", |
| "paper_id": "2025PhRvD.111f3523P", |
| "question": "For constraining matter density and clustering from simulated weak-lensing measurements with observational noise, which dimensionality-reduction method gives tighter posterior constraints?", |
| "option_a": "neural-network compression trained to predict cosmological parameters", |
| "option_b": "canonical correlation analysis", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0244", |
| "paper_id": "2025PhRvD.111f3523P", |
| "question": "For constraining matter density and clustering from simulated weak-lensing second moments, which dimensionality-reduction method gives tighter posterior constraints?", |
| "option_a": "neural-network compression trained to predict cosmological parameters", |
| "option_b": "principal component analysis", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0245", |
| "paper_id": "2025A&A...694A.326M", |
| "question": "For estimating stellar temperature and surface gravity from observed medium-resolution spectra, which ML pipeline gives lower errors?", |
| "option_a": "multi-discriminator adversarial autoencoder + ANN", |
| "option_b": "standard autoencoder + ANN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0246", |
| "paper_id": "2025ApJ...981L..24L", |
| "question": "For constraining cosmic curvature from baryon acoustic oscillation and sparsely sampled cosmic chronometer measurements, which reconstruction method yields tighter constraints?", |
| "option_a": "artificial neural network", |
| "option_b": "Gaussian process", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0247", |
| "paper_id": "2025ApJ...992..152R", |
| "question": "For identifying binary black hole mergers from gravitational-wave alerts, which neural-network pipeline gives higher classification accuracy?", |
| "option_a": "multiclass MLP using alert properties", |
| "option_b": "one-vs-all CNNs using localization images", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0248", |
| "paper_id": "2025ApJ...992..152R", |
| "question": "For classifying gravitational-wave alerts, which supervised method better retains its accuracy when moving from simulated mergers to real alerts?", |
| "option_a": "multilayer perceptron", |
| "option_b": "XGBoost", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0249", |
| "paper_id": "2025AJ....170...73F", |
| "question": "For separating planet candidates from false positives among Kepler transit detections, which CNN pipeline gives a higher F1 score?", |
| "option_a": "CNN using phase-folded light curves alone", |
| "option_b": "CNN using light curves and diagnostic features", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0250", |
| "paper_id": "2025ApJ...989...47L", |
| "question": "For inferring the amplitude of matter fluctuations from simulated dark-matter halo catalogs, which neural-network architecture is more accurate?", |
| "option_a": "topological neural network", |
| "option_b": "graph neural network", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0251", |
| "paper_id": "2025AJ....170..263H", |
| "question": "For distinguishing self-interacting from cold dark matter in galaxy clusters using simulated maps without observational noise, which neural-network architecture achieves comparable classification accuracy with fewer trainable parameters?", |
| "option_a": "convolutional Kolmogorov-Arnold network", |
| "option_b": "Inception CNN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0252", |
| "paper_id": "2025ApJS..278...54Y", |
| "question": "For reconstructing extended radio sources from radio-interferometric observations, which method gives lower reconstruction error?", |
| "option_a": "split-Bregman network with a fixed wavelet dictionary", |
| "option_b": "split-Bregman network with a learned deep dictionary", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0253", |
| "paper_id": "2025ApJ...984..106N", |
| "question": "For identifying LINER regions in galaxy spectra after removing the emission lines, which method gives a higher detection rate?", |
| "option_a": "attention-based CNN classifier", |
| "option_b": "PCA + logistic regression", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0254", |
| "paper_id": "2025ApJ...990...47P", |
| "question": "For finding strong gravitational lenses in observed images with models trained on simulated lenses, which unsupervised domain-adaptation method recovers more lenses at low false-positive rates?", |
| "option_a": "Wasserstein distance-guided domain adaptation", |
| "option_b": "adversarial discriminative domain adaptation", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0255", |
| "paper_id": "2025ApJ...990...47P", |
| "question": "For finding strong gravitational lenses in observed images with models trained on simulated lenses, which ML pipeline recovers more lenses at low false-positive rates?", |
| "option_a": "residual CNN + Wasserstein domain adaptation", |
| "option_b": "equivariant CNN + Wasserstein domain adaptation", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0256", |
| "paper_id": "2025ApJS..278...39Z", |
| "question": "For classifying galaxy morphology from observed images, which ML pipeline gives higher accuracy and recall?", |
| "option_a": "convolutional-autoencoder encoding + ensemble clustering", |
| "option_b": "ConvNeXt-Vision Transformer contrastive encoding + ensemble clustering", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0257", |
| "paper_id": "2025ApJS..278...18Z", |
| "question": "For selecting nearby dwarf-galaxy candidates for spectroscopic follow-up using broadband photometry, which supervised method gives higher precision?", |
| "option_a": "multilayer perceptron", |
| "option_b": "XGBoost", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0258", |
| "paper_id": "2025AJ....169..282J", |
| "question": "For detecting fast-moving objects in astronomical images where stars and moving objects may appear either point-like or trailed, which method gives better overall detection performance?", |
| "option_a": "gated mixture of physics-informed object detectors", |
| "option_b": "single physics-informed object detector", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0259", |
| "paper_id": "2025AJ....169..282J", |
| "question": "For detecting fast-moving objects in astronomical images with bright sky backgrounds, which method maintains higher recall?", |
| "option_a": "physics-informed center-point object detector", |
| "option_b": "image-only center-point object detector", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0260", |
| "paper_id": "2025PhRvD.111l3519Z", |
| "question": "For emulating weak-lensing correlation functions over a broad range of cosmological parameters, which method gives lower prediction error?", |
| "option_a": "residual multilayer perceptron", |
| "option_b": "self-attention Transformer", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0261", |
| "paper_id": "2025AJ....170..255H", |
| "question": "For reconstructing multiband stellar photometry, which method gives lower reconstruction error?", |
| "option_a": "standard autoencoder", |
| "option_b": "LSTM autoencoder", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0262", |
| "paper_id": "2025PhRvD.112h3510B", |
| "question": "For detecting damped Lyman-alpha absorbers in simulated quasar spectra affected by broad absorption lines, which ML method is less susceptible to false detections?", |
| "option_a": "Gaussian-process model", |
| "option_b": "convolutional neural network", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0263", |
| "paper_id": "2025PhRvD.111l3520S", |
| "question": "For emulating weak-lensing correlation functions over a broad range of cosmological parameters, which neural-network architecture gives lower prediction error?", |
| "option_a": "hybrid residual-MLP/self-attention Transformer", |
| "option_b": "self-attention-only Transformer", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0264", |
| "paper_id": "2025A&A...699A..54S", |
| "question": "For detecting Ellerman bombs in H-alpha observations at reduced spatial resolution, which neural-network architecture gives a higher F1 score?", |
| "option_a": "convolutional neural network", |
| "option_b": "fully connected neural network", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0265", |
| "paper_id": "2025PASP..137c4507S", |
| "question": "For distinguishing hydrogen-rich from hydrogen-poor core-collapse supernovae in low-resolution spectra, which deep-learning approach achieves higher recall at a low false-positive rate?", |
| "option_a": "multiclass convolutional classifier", |
| "option_b": "parallel binary recurrent classifiers", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0266", |
| "paper_id": "2025PASP..137c4507S", |
| "question": "When high-quality low-resolution spectra are available, which neural-network pipeline more accurately distinguishes hydrogen-rich from hydrogen-poor core-collapse supernovae?", |
| "option_a": "bidirectional LSTM with additional light-curve branches", |
| "option_b": "bidirectional LSTM using spectra", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0267", |
| "paper_id": "2025A&A...694A.287B", |
| "question": "For emulating the matter distribution in cosmological N-body simulations, which ML pipeline gives lower density-field errors?", |
| "option_a": "direct neural displacement-field emulator", |
| "option_b": "neural reference-frame emulator with force corrections", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0268", |
| "paper_id": "2025ApJ...993...33S", |
| "question": "For classifying radio-background models from cosmic-dawn 21 cm observations, which neural-network pipeline gives higher accuracy?", |
| "option_a": "single-layer sigmoid MLP on the global 21 cm signal", |
| "option_b": "three-layer ReLU MLP on the 21 cm power spectrum", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0269", |
| "paper_id": "2025A&C....52..965L", |
| "question": "For identifying spiral galaxies with at least four arms as one class, which transfer-learning pipeline is more accurate?", |
| "option_a": "frozen EfficientNetB0 CNN", |
| "option_b": "fully fine-tuned EfficientNetV2 CNN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0270", |
| "paper_id": "2025A&A...697A.148B", |
| "question": "For classifying emission-line regions whose spectra span a broad radial-velocity range, which neural-network architecture maintains higher F1 scores?", |
| "option_a": "one-dimensional convolutional neural network", |
| "option_b": "fully connected neural network", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0271", |
| "paper_id": "2025A&A...698A..80M", |
| "question": "For predicting simulated 21 cm power spectra from reionization parameters, which neural-network approach gives lower prediction error?", |
| "option_a": "single neural-network emulator", |
| "option_b": "ensemble of independently trained neural-network emulators", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0272", |
| "paper_id": "2025A&A...698A..80M", |
| "question": "For inferring reionization parameters from noisy simulated 21 cm power spectra on a coarsely sampled parameter grid, which ML pipeline gives more accurate and better calibrated posterior estimates?", |
| "option_a": "neural-network emulator with MCMC", |
| "option_b": "neural density estimator with MCMC", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0273", |
| "paper_id": "2025A&C....5100929B", |
| "question": "For compressing observed solar polarization spectra when the proportions of quiet-Sun and active-region profiles vary across training sets, which autoencoder architecture gives more stable reconstruction errors?", |
| "option_a": "fully connected autoencoder", |
| "option_b": "one-dimensional convolutional autoencoder", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0274", |
| "paper_id": "2025A&A...694A..88R", |
| "question": "For estimating exoplanet atmospheric mass-loss rates from a grid of hydrodynamic models, which method gives more accurate predictions?", |
| "option_a": "radial basis function regression", |
| "option_b": "fully connected neural network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0275", |
| "paper_id": "2025ApJ...992..156A", |
| "question": "For separating gamma-ray-induced air showers from hadronic background at intermediate energies, which method gives stronger background rejection?", |
| "option_a": "single-branch MLP using all event features", |
| "option_b": "two-branch CNN-MLP with radial charge features processed separately", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0276", |
| "paper_id": "2025ApJS..276...20Z", |
| "question": "For detecting fast radio bursts in follow-up radio observations of sources with known dispersion measures, which deep-learning approach recovers more bursts?", |
| "option_a": "ResNet binary classifier", |
| "option_b": "CenterNet object detector", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0277", |
| "paper_id": "2025AJ....169...24C", |
| "question": "For recovering single-epoch injected radial-velocity shifts from high-resolution stellar spectra affected by stellar variability, which approach yields lower spectrum-level prediction error?", |
| "option_a": "linear ensemble of single-line CNNs", |
| "option_b": "single-line CNN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0278", |
| "paper_id": "2025ApJ...981..157L", |
| "question": "For identifying the magnetic class of solar active regions from magnetogram cutouts, which transformer architecture is more reliable across the classes?", |
| "option_a": "Vision Transformer (ViT Base)", |
| "option_b": "Data-efficient Image Transformer (DeiT Base)", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0279", |
| "paper_id": "2025ApJ...981..157L", |
| "question": "For identifying the magnetic class of solar active regions from paired magnetogram and continuum cutouts, which CNN family is more reliable across the classes?", |
| "option_a": "ResNet", |
| "option_b": "Inception", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0280", |
| "paper_id": "2025PASP..137e4503I", |
| "question": "For detecting fast-moving-object streaks in wide-field difference images, which pipeline achieves higher detection completeness while maintaining very high precision?", |
| "option_a": "thresholded linear-candidate extraction + deep-classifier + human-vetting pipeline", |
| "option_b": "encoder-decoder CNN segmentation + endpoint-fitting pipeline", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0281", |
| "paper_id": "2025A&A...697A.107Y", |
| "question": "For identifying carbon stars from low-resolution spectra, which method achieves higher recall?", |
| "option_a": "XGBoost", |
| "option_b": "multi-scale one-dimensional CNN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0282", |
| "paper_id": "2025A&A...695A..81U", |
| "question": "For identifying pulsating components in eclipsing-binary light curves, which object detector processes the light curves faster while giving higher overall detection accuracy?", |
| "option_a": "YOLO", |
| "option_b": "EfficientDet", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0283", |
| "paper_id": "2025A&C....5100935T", |
| "question": "For emulating stellar density profiles in binary-population simulations, which method requires less storage while retaining comparable profile accuracy?", |
| "option_a": "nearest-neighbor profile approximation", |
| "option_b": "PCA + feed-forward neural-network emulator", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0284", |
| "paper_id": "2025PhRvD.111d1301Y", |
| "question": "For reconstructing the arrival directions of simulated muon-neutrino events in a neutrino telescope, which deep-learning pipeline gives finer angular resolution?", |
| "option_a": "VAE-UNet++ super-resolution followed by a sparse CNN", |
| "option_b": "sparse CNN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0285", |
| "paper_id": "2025ApJ...988..204C", |
| "question": "For determining the bolometric luminosities of type 1 quasars from multi-band photometry, which ML approach is more accurate?", |
| "option_a": "self-organizing map", |
| "option_b": "multiple linear regression", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0286", |
| "paper_id": "2025A&A...696A.214P", |
| "question": "For identifying strong gravitational lenses in real survey images, which EfficientNet training strategy gives better classification performance?", |
| "option_a": "EfficientNet trained from scratch for lens classification", |
| "option_b": "morphology-pretrained EfficientNet fine-tuned for lens classification", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0287", |
| "paper_id": "2025A&A...696A.214P", |
| "question": "For identifying strong gravitational lenses in real survey images, which CNN architecture gives better classification performance?", |
| "option_a": "U-Net", |
| "option_b": "Mask R-CNN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0288", |
| "paper_id": "2025MNRAS.541.1682C", |
| "question": "For reproducing the distribution of stellar metallicities in central galaxies in dark-matter-only simulations, which neural-network pipeline is more accurate?", |
| "option_a": "stochastic semi-recurrent neural network", |
| "option_b": "deterministic semi-recurrent neural network", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0289", |
| "paper_id": "2025A&A...693A.170D", |
| "question": "For reconstructing a simulated chromospheric line-of-sight magnetic-field map, which neural-network parameterization gives lower error?", |
| "option_a": "Fourier-feature neural field", |
| "option_b": "ReLU neural field", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0290", |
| "paper_id": "2025ApJS..281...47Y", |
| "question": "For removing Galactic foregrounds from simulated cosmic microwave background polarization maps with instrumental noise, which approach gives lower map-reconstruction error?", |
| "option_a": "Transformer on spherical maps", |
| "option_b": "convolutional neural network on flat map projections", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0291", |
| "paper_id": "2025ApJS..281...47Y", |
| "question": "For removing Galactic foregrounds from simulated cosmic microwave background polarization maps, which approach is less expensive to train?", |
| "option_a": "convolutional neural network on flat map projections", |
| "option_b": "Transformer on spherical maps", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0292", |
| "paper_id": "2025MNRAS.537..640Y", |
| "question": "For classifying bars when a galaxy-morphology model is applied to images from another survey, which CNN training strategy gives higher accuracy?", |
| "option_a": "source-trained CNN applied without adaptation", |
| "option_b": "CNN adapted with unlabeled images from the new survey", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0293", |
| "paper_id": "2025ApJS..277...10G", |
| "question": "For determining whether an astronomical image contains a galaxy, which classifier gives more reliable classifications?", |
| "option_a": "one-dimensional CNN on Zernike-moment features", |
| "option_b": "support-vector machine on Zernike-moment features", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0294", |
| "paper_id": "2025ApJS..277...10G", |
| "question": "For classifying galaxies as spiral, elliptical, or peculiar, which ML pipeline gives more balanced classifications?", |
| "option_a": "support-vector machine on Zernike-moment features", |
| "option_b": "shallow CNN on galaxy images", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0295", |
| "paper_id": "2025ApJS..277...10G", |
| "question": "For classifying galaxies as spiral, elliptical, or peculiar, which CNN architecture gives more accurate classifications?", |
| "option_a": "shallow CNN", |
| "option_b": "VGG16 CNN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0296", |
| "paper_id": "2025A&A...699A.350S", |
| "question": "For identifying strong gravitational lenses while keeping false positives very low, which approach recovers more lenses?", |
| "option_a": "single residual CNN", |
| "option_b": "ensemble of residual CNNs", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0297", |
| "paper_id": "2025A&A...699A.350S", |
| "question": "For improving the purity of a strong-lens candidate sample without reducing lens completeness, which ML pipeline is more effective?", |
| "option_a": "residual-CNN ensemble followed by a lens-modeling CNN", |
| "option_b": "residual-CNN ensemble", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0298", |
| "paper_id": "2025A&A...698A.276J", |
| "question": "For estimating photometric redshifts of bright galaxies, which ML pipeline gives more precise estimates?", |
| "option_a": "CNN using galaxy images combined with an ANN using magnitudes", |
| "option_b": "ANN using magnitudes", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0299", |
| "paper_id": "2025ApJS..279...47L", |
| "question": "For identifying unresolved white-dwarf-main-sequence binaries when one star dominates the optical spectrum, which ML pipeline recovers more systems?", |
| "option_a": "neural spectral fitting with a Gaussian-process classifier", |
| "option_b": "neural spectral fitting alone", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0300", |
| "paper_id": "2025ApJS..279...28S", |
| "question": "For identifying ion-kinetic instabilities when training and evaluation samples come from the same solar-wind simulations, which ML pipeline gives more reliable classifications?", |
| "option_a": "CNN on velocity-distribution images", |
| "option_b": "random forest on velocity-distribution moments", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0301", |
| "paper_id": "2025PASP..137d4503S", |
| "question": "For identifying eclipsing binaries from light curves, which neural-network architecture gives more accurate classifications?", |
| "option_a": "two-branch CNN", |
| "option_b": "two-branch CNN-attention-LSTM", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0302", |
| "paper_id": "2025ApJ...988...59F", |
| "question": "For reconstructing irregularly sampled multiband quasar light curves in simulations, which ML method gives better probabilistic reconstructions?", |
| "option_a": "multitask Gaussian-process regression with a damped-random-walk kernel", |
| "option_b": "latent stochastic-differential-equation neural network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0303", |
| "paper_id": "2025A&A...701A.150H", |
| "question": "For classifying variable sources using low-resolution spectra encoded by an autoencoder, which classifier gives better overall performance?", |
| "option_a": "logistic regression", |
| "option_b": "k-nearest-neighbor classifier", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0304", |
| "paper_id": "2025ApJ...993...58F", |
| "question": "For estimating photospheric magnetic-field strength from line-of-sight magnetograms, which U-Net formulation is more accurate?", |
| "option_a": "U-Net that directly regresses the magnetic-field components with L1 loss", |
| "option_b": "U-Net that classifies and refines magnetic-field strength and direction", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0305", |
| "paper_id": "2025AJ....169..159V", |
| "question": "For identifying microlensing events from light curves observed at multiple sites, which ML pipeline is more accurate?", |
| "option_a": "multi-branch recurrent neural network", |
| "option_b": "XGBoost on fitted light-curve features", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0306", |
| "paper_id": "2025ApJ...994..110K", |
| "question": "For distinguishing binary from single stars using light curves and spectra, which multimodal method is more accurate?", |
| "option_a": "self- and cross-attention with dual-projection alignment", |
| "option_b": "variance-invariance-covariance regularized alignment", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0307", |
| "paper_id": "2025ApJ...983..185F", |
| "question": "For estimating black-hole spin from simulated photon-ring images at lower angular resolution than the training images, which ML pipeline gives more accurate estimates?", |
| "option_a": "convolutional neural network on full photon-ring images", |
| "option_b": "gradient-boosted regression trees on extracted photon-ring geometry", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0308", |
| "paper_id": "2025MNRAS.544.1514A", |
| "question": "For selecting galaxies within the colour space covered by a spectroscopic training sample, which classifier gives a higher AUC?", |
| "option_a": "random forest classifier", |
| "option_b": "k-nearest-neighbor classifier", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0309", |
| "paper_id": "2025MNRAS.544.1514A", |
| "question": "For estimating photometric redshifts of higher-redshift galaxies from broadband photometry, which regression method gives lower scatter?", |
| "option_a": "feed-forward neural network", |
| "option_b": "Gaussian-process regression", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0310", |
| "paper_id": "2025A&A...694A.212B", |
| "question": "For classifying observed ionised nebulae using labelled photoionisation simulations for training, which neural-network architecture gives higher accuracy?", |
| "option_a": "feed-forward neural network", |
| "option_b": "domain-adversarial neural network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0311", |
| "paper_id": "2025ApJ...980..245C", |
| "question": "For estimating intracluster light fractions from galaxy-cluster images, which ML pipeline gives lower prediction error?", |
| "option_a": "CNN trained on injected intracluster-light profiles without real-data fine-tuning", |
| "option_b": "CNN pretrained on images with injected intracluster-light profiles and fine-tuned on real cluster images", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0312", |
| "paper_id": "2025ApJS..281...27P", |
| "question": "For determining the surface gravities of dwarf stars from high-resolution spectra and photometry, which method is more accurate?", |
| "option_a": "neural spectral model with stellar-evolution constraints", |
| "option_b": "neural spectral model without stellar-evolution constraints", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0313", |
| "paper_id": "2025ApJS..279...26F", |
| "question": "For classifying stars, quasars, and galaxies from multiband observations, which ML pipeline gives more accurate classifications?", |
| "option_a": "CNN-MLP combining source images and multiband photometry", |
| "option_b": "MLP using multiband photometry", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0314", |
| "paper_id": "2025ApJS..279...26F", |
| "question": "For classifying stars, quasars, and galaxies from multiband photometry, which method gives more accurate classifications?", |
| "option_a": "XGBoost", |
| "option_b": "multilayer perceptron", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0315", |
| "paper_id": "2025A&C....5300998P", |
| "question": "For classifying detached and overcontact eclipsing binaries from survey light curves, which neural-network architecture gives higher accuracy across different cadences?", |
| "option_a": "vision Transformer", |
| "option_b": "residual convolutional neural network", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0316", |
| "paper_id": "2025A&A...694A.183B", |
| "question": "For classifying variable sources from multiband light curves, which recurrent-network architecture gives more accurate classifications?", |
| "option_a": "band-specific LSTMs followed by direct feature concatenation", |
| "option_b": "band-specific LSTMs followed by a central LSTM", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0317", |
| "paper_id": "2025A&A...694A.183B", |
| "question": "For classifying variable stars from multiband light curves with few observations in each band, which ML method gives more accurate classifications?", |
| "option_a": "multiband recurrent neural network", |
| "option_b": "balanced random forest using light-curve features", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0318", |
| "paper_id": "2025ApJS..279...49G", |
| "question": "For simulating nucleosynthesis during the late burning stages of massive stars, which neural-network design achieves similar accuracy with shorter total inference time?", |
| "option_a": "one network for isotopic abundances and another for nuclear energy generation", |
| "option_b": "single network jointly predicting isotopic abundances and nuclear energy generation", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0319", |
| "paper_id": "2025MNRAS.540.3753B", |
| "question": "For reproducing the spread in the stellar mass-metallicity relation of simulated intermediate- and high-mass galaxies, which ML pipeline is more accurate?", |
| "option_a": "RNN with an auxiliary Fourier-amplitude network and variance-normalized stochastic corrections", |
| "option_b": "RNN predicting star-formation and metallicity histories directly", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0320", |
| "paper_id": "2025PhRvD.111j4029T", |
| "question": "For predicting the remnant properties of merging binary black holes at comparable accuracy, which machine-learning method gives faster evaluations?", |
| "option_a": "Gaussian process regression", |
| "option_b": "fully connected neural network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0321", |
| "paper_id": "2025A&A...700A..87S", |
| "question": "For estimating dark-matter halo formation times from simulated galaxy clusters, which machine-learning pipeline gives estimates with less scatter?", |
| "option_a": "convolutional network using radial gas and stellar property maps", |
| "option_b": "random forest using halo and galaxy properties", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0322", |
| "paper_id": "2025ApJ...992..153N", |
| "question": "For identifying quasars from survey light curves, which ML pipeline gives a higher F1 score?", |
| "option_a": "XGBoost using precomputed light-curve features", |
| "option_b": "Transformer trained directly on light curves", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0323", |
| "paper_id": "2025MNRAS.538.1096R", |
| "question": "For simulating X-ray reflection spectra of active galactic nuclei, which neural-network approach gives lower prediction error?", |
| "option_a": "single feed-forward neural network", |
| "option_b": "ensemble of feed-forward neural networks", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0324", |
| "paper_id": "2025MNRAS.538.2905M", |
| "question": "For generating low-frequency radio images of galaxy-cluster halos, which generative model produces more realistic images?", |
| "option_a": "Wasserstein generative adversarial network", |
| "option_b": "denoising diffusion model", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0325", |
| "paper_id": "2025MNRAS.538.2905M", |
| "question": "For identifying central diffuse emission in low-frequency radio images of galaxy clusters, which neural-network architecture gives higher classification accuracy?", |
| "option_a": "convolutional neural network with multi-head attention", |
| "option_b": "conventional convolutional neural network", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0326", |
| "paper_id": "2025A&A...696A.114P", |
| "question": "For constraining Galactic radio-pulsar population parameters from simulations, which neural inference strategy reaches comparable results with fewer simulations?", |
| "option_a": "amortized neural posterior estimation", |
| "option_b": "truncated sequential neural posterior estimation", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0327", |
| "paper_id": "2025AJ....169..304T", |
| "question": "For classifying transients from the first few days of simulated light curves, which ML approach gives higher overall accuracy?", |
| "option_a": "semi-supervised variational autoencoder with a recurrent classifier", |
| "option_b": "supervised recurrent neural network", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0328", |
| "paper_id": "2025ApJS..280...50J", |
| "question": "For reconstructing solar extreme-ultraviolet irradiance from Ca II K images, which neural-network architecture gives lower prediction error?", |
| "option_a": "vision Transformer", |
| "option_b": "residual convolutional neural network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0329", |
| "paper_id": "2025MNRAS.538.1081R", |
| "question": "For reducing false positives in strong-lens searches while retaining most lenses, which CNN strategy is more effective?", |
| "option_a": "single EfficientNet CNN", |
| "option_b": "DenseNet-EfficientNet ensemble", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0330", |
| "paper_id": "2025ApJS..281...58L", |
| "question": "For estimating atmospheric parameters of M-type stars from low-resolution spectra, which ML pipeline gives more precise estimates?", |
| "option_a": "support-vector regression", |
| "option_b": "convolutional neural-network ensemble", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0331", |
| "paper_id": "2025ApJS..281...58L", |
| "question": "For estimating atmospheric parameters of M-type stars, which ML pipeline gives more precise estimates?", |
| "option_a": "PCA-LightGBM regression from photometry", |
| "option_b": "CNN regression from low-resolution spectra", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0332", |
| "paper_id": "2025A&A...703A.100N", |
| "question": "For estimating open-cluster metallicities, which machine-learning approach gives lower error?", |
| "option_a": "XGBoost using low-resolution spectra and broadband photometry", |
| "option_b": "neural network using color-magnitude-diagram features", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0333", |
| "paper_id": "2025A&A...693A.245W", |
| "question": "For distinguishing hot subdwarfs from O stars and white dwarfs using photometric data, which ML pipeline gives a higher F1 score?", |
| "option_a": "magnitude-based CatBoost", |
| "option_b": "image-based Swin Transformer", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0334", |
| "paper_id": "2025A&A...693A.245W", |
| "question": "For distinguishing hot subdwarfs from O stars and white dwarfs in multiband images, which Transformer architecture gives a higher F1 score?", |
| "option_a": "shifted-window vision Transformer", |
| "option_b": "standard vision Transformer", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0335", |
| "paper_id": "2025A&A...699A.179D", |
| "question": "For fitting simulated high-resolution X-ray spectra with a limited simulation budget, which normalizing-flow pipeline produces more accurate posterior distributions?", |
| "option_a": "normalizing flow with multilayer-perceptron compression", |
| "option_b": "normalizing flow with predefined spectral summaries", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0336", |
| "paper_id": "2025MNRAS.536..962N", |
| "question": "For estimating the cold neutral hydrogen fraction and opacity correction from simulated 21 cm emission spectra, which neural-network architecture gives lower prediction error?", |
| "option_a": "deep convolutional network", |
| "option_b": "convolutional-Transformer network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0337", |
| "paper_id": "2025ApJ...980..108B", |
| "question": "For reconstructing simulated strongly lensed galaxies when the training population does not match the source population, which score-based inference strategy gives less biased reconstructions?", |
| "option_a": "score-based inference with an iteratively updated prior", |
| "option_b": "score-based inference with a fixed prior", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0338", |
| "paper_id": "2025MNRAS.537.1608T", |
| "question": "For reconstructing simulated high-dynamic-range radio-interferometric images, which neural-network approach gives higher accuracy across bright and faint emission?", |
| "option_a": "end-to-end U-Net reconstruction", |
| "option_b": "constrained plug-and-play reconstruction with a neural-network denoiser", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0339", |
| "paper_id": "2025A&A...703A.301B", |
| "question": "For estimating cosmological parameters from simulated matter power spectra, which regression model gives lower prediction error?", |
| "option_a": "multilayer perceptron", |
| "option_b": "linear regression", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0340", |
| "paper_id": "2025ApJS..279...54F", |
| "question": "For estimating quasar redshifts from multiband photometry, which ML approach gives lower scatter in the redshift errors?", |
| "option_a": "ensemble of XGBoost, TabNet, and FT-Transformer", |
| "option_b": "TabNet", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0341", |
| "paper_id": "2025ApJ...991..152D", |
| "question": "For predicting the global 21 cm signal from cosmological and astrophysical parameters, which neural-network architecture achieves similar accuracy with much less training time?", |
| "option_a": "long short-term memory network", |
| "option_b": "Kolmogorov-Arnold network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0342", |
| "paper_id": "2025ApJ...991..152D", |
| "question": "For predicting the global 21 cm signal from cosmological and astrophysical parameters, which neural-network architecture gives lower prediction error and requires less training time?", |
| "option_a": "Kolmogorov-Arnold network", |
| "option_b": "fully connected neural network", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0343", |
| "paper_id": "2025ApJ...987...96M", |
| "question": "For estimating the mass of a dark-matter subhalo from a simulated stellar stream, which ML pipeline provides tighter and better-calibrated constraints?", |
| "option_a": "graph convolutional network using individual stellar positions and velocities", |
| "option_b": "dense neural network using the stellar-density power spectrum", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0344", |
| "paper_id": "2025ApJ...991..120J", |
| "question": "For estimating the amplitude of matter fluctuations from neutral-hydrogen maps across different cosmological simulation models, which neural-network approach gives less biased predictions?", |
| "option_a": "domain-invariant CNN", |
| "option_b": "standard CNN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0345", |
| "paper_id": "2025A&A...699A.168C", |
| "question": "For identifying incorrect variable-star classifications from light curves, which Transformer-based approach provides more reliable uncertainty estimates?", |
| "option_a": "hierarchical-attention Transformer with Monte Carlo dropout", |
| "option_b": "ensemble of independently trained Transformer classifiers", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0346", |
| "paper_id": "2025A&A...702A.148M", |
| "question": "For estimating RRab metallicities from Gaia G-band light curves, which deep-learning pipeline gives lower prediction error?", |
| "option_a": "bidirectional LSTM pipeline with binned light curves", |
| "option_b": "GRU pipeline with spline-interpolated light curves", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0347", |
| "paper_id": "2025PhRvD.111h3056P", |
| "question": "For estimating the surface shape of rotating neutron stars across different equations of state, which regression pipeline gives lower maximum relative error?", |
| "option_a": "polynomial surface model", |
| "option_b": "physically constrained feed-forward neural-network surface model", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0348", |
| "paper_id": "2025PhRvD.112b3503C", |
| "question": "For detecting parity violation in simulated two-dimensional galaxy distributions, which unsupervised method requires fewer training examples to reach statistical significance?", |
| "option_a": "trainable-filter scattering transform", |
| "option_b": "convolutional neural network", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0349", |
| "paper_id": "2025PhRvD.112b3503C", |
| "question": "For detecting parity violation in simulated two-dimensional galaxy distributions, which unsupervised method requires fewer training examples to reach statistical significance?", |
| "option_a": "fixed-wavelet scattering transform", |
| "option_b": "trainable-filter scattering transform", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0350", |
| "paper_id": "2025A&C....5200953P", |
| "question": "For reconstructing lunar terrain from simulated orbital images, which 3D reconstruction pipeline gives lower geometric error?", |
| "option_a": "photogrammetry combined with 3D Gaussian splatting", |
| "option_b": "neural radiance field", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0351", |
| "paper_id": "2025A&C....5200953P", |
| "question": "For reconstructing lunar terrain from simulated orbital images, which neural 3D reconstruction method produces smaller maximum surface deviations?", |
| "option_a": "neural radiance field", |
| "option_b": "3D Gaussian splatting", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0352", |
| "paper_id": "2025AJ....169..146C", |
| "question": "For detecting faint sources in astronomical images, which deep-learning pipeline gives higher precision, recall, and F1 score?", |
| "option_a": "region-based detection CNN", |
| "option_b": "densely nested segmentation CNN with bit-depth image decomposition and soft supervision", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0353", |
| "paper_id": "2025AJ....170...21H", |
| "question": "For vetting transiting exoplanet candidates from light curves, which architecture gives higher classification accuracy?", |
| "option_a": "CNN-LSTM without attention", |
| "option_b": "CNN-LSTM with additive attention", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0354", |
| "paper_id": "2025AJ....170..158Z", |
| "question": "For identifying blue horizontal-branch stars from low-resolution spectra, which architecture gives higher classification accuracy and F1 score?", |
| "option_a": "CNN with attention and GRU layers", |
| "option_b": "ResNet-18 CNN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0355", |
| "paper_id": "2025AJ....170..158Z", |
| "question": "For identifying blue horizontal-branch stars from low-resolution spectra, which training strategy gives higher classification accuracy and F1 score?", |
| "option_a": "binary-classifier training without transfer learning", |
| "option_b": "transfer learning with frozen feature layers", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0356", |
| "paper_id": "2025AJ....170..207U", |
| "question": "For identifying interstellar filaments and measuring their position angles, which approach better preserves filament morphology?", |
| "option_a": "traditional image-processing filament extraction + U-Net angle classification", |
| "option_b": "Mask R-CNN filament extraction + U-Net angle regression", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0357", |
| "paper_id": "2025ApJ...978..128H", |
| "question": "For identifying the open magnetic-field footpoints associated with coronal holes in magnetohydrodynamic simulations, which classifier is more accurate?", |
| "option_a": "linear support-vector machine", |
| "option_b": "multilayer perceptron", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0358", |
| "paper_id": "2025ApJ...978..151L", |
| "question": "For detecting strong gravitational lenses in galaxy images, which YOLOX architecture gives higher precision?", |
| "option_a": "YOLOX with selective-kernel attention", |
| "option_b": "YOLOX without attention", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0359", |
| "paper_id": "2025ApJ...978..151L", |
| "question": "For detecting strong gravitational lenses in galaxy-centered image cutouts, which bounding-box loss gives more accurate detections?", |
| "option_a": "intersection-over-union loss", |
| "option_b": "generalized intersection-over-union loss", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0360", |
| "paper_id": "2025ApJ...984...67S", |
| "question": "For forecasting strongly southward interplanetary magnetic-field Bz over the next six hours, which neural-network architecture gives lower prediction error?", |
| "option_a": "bidirectional LSTM", |
| "option_b": "multilayer perceptron", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0361", |
| "paper_id": "2025ApJ...984...87V", |
| "question": "For forecasting the maximum solar X-ray flux over the next day from magnetic-field and flare-history measurements across multiple solar cycles, which regression method gives lower prediction error?", |
| "option_a": "linear regression", |
| "option_b": "random forest regression", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0362", |
| "paper_id": "2025ApJ...988..118H", |
| "question": "For forecasting the evolution of geomagnetic activity over a solar cycle, which method is more accurate?", |
| "option_a": "LSTM with iterative forecast feedback", |
| "option_b": "feed-forward neural network", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0363", |
| "paper_id": "2025ApJ...993..146Y", |
| "question": "For extrapolating coronal magnetic fields from photospheric magnetograms, which method better preserves the divergence-free condition?", |
| "option_a": "Fourier neural operator with frequency-domain physical constraints", |
| "option_b": "Fourier neural operator with finite-difference physical constraints", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0364", |
| "paper_id": "2025ApJS..276....4L", |
| "question": "For detecting point sources in astronomical images, which neural-network pipeline achieves higher purity and completeness?", |
| "option_a": "U-Net with Laplacian-of-Gaussian localization", |
| "option_b": "residual encoder-decoder with feature-template matching", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0365", |
| "paper_id": "2025ApJS..276...24L", |
| "question": "For separating simulated gamma-ray-induced air showers from proton background, which machine-learning approach gives more accurate classifications?", |
| "option_a": "stacking ensemble of classifiers", |
| "option_b": "XGBoost classifier", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0366", |
| "paper_id": "2025ApJS..276...37L", |
| "question": "For identifying Ca II absorption lines in quasar spectra, which CNN architecture gives more accurate classifications?", |
| "option_a": "residual CNN", |
| "option_b": "residual CNN with channel and spatial attention", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0367", |
| "paper_id": "2025ApJS..276...39L", |
| "question": "For identifying pulsar candidates after updating a classifier with data from another radio telescope, which neural-network approach better retains its performance on earlier observations?", |
| "option_a": "Bayesian convolutional neural network", |
| "option_b": "deterministic convolutional neural network", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0368", |
| "paper_id": "2025ApJS..276...68X", |
| "question": "For forecasting C-class or stronger solar flares from active-region magnetic fields, which neural-network approach produces fewer false alarms?", |
| "option_a": "CNN", |
| "option_b": "CNN-TCN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0369", |
| "paper_id": "2025ApJS..277...54J", |
| "question": "For extrapolating active-region coronal magnetic fields from photospheric vector magnetograms, which neural-operator approach better satisfies the divergence-free condition?", |
| "option_a": "physics-informed neural operator", |
| "option_b": "data-driven neural operator", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0370", |
| "paper_id": "2025ApJS..278...35M", |
| "question": "For classifying the morphologies of interacting galaxies from low-resolution images, which superresolution model leads to more accurate classifications?", |
| "option_a": "dual-attention Transformer GAN", |
| "option_b": "dual-branch CNN-Transformer GAN", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0371", |
| "paper_id": "2025ApJS..278...63L", |
| "question": "For forecasting M-class or stronger solar flares within 24 hours from active-region magnetograms, which ML approach gives a higher true skill score?", |
| "option_a": "convolutional neural network", |
| "option_b": "multiscale vision Transformer", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0372", |
| "paper_id": "2025ApJS..279...23Y", |
| "question": "For segmenting coronal mass ejections in coronagraph images, which method produces more accurate segmentations?", |
| "option_a": "CNN segmentation network", |
| "option_b": "CNN-Transformer segmentation network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0373", |
| "paper_id": "2025ApJS..279...23Y", |
| "question": "For segmenting coronal mass ejections in coronagraph images, which method produces more accurate segmentations?", |
| "option_a": "Transformer segmentation network", |
| "option_b": "CNN-Transformer segmentation network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0374", |
| "paper_id": "2025ApJS..279...24L", |
| "question": "For detecting interplanetary coronal mass ejections in solar-wind time series, which deep-learning pipeline misses fewer events?", |
| "option_a": "CNN detection pipeline", |
| "option_b": "CNN-LSTM detection pipeline", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0375", |
| "paper_id": "2025ApJS..279...36Z", |
| "question": "For classifying stars from low-resolution spectra, which deep-learning pipeline gives higher overall accuracy?", |
| "option_a": "one-dimensional CNN + ResNet ensemble", |
| "option_b": "one-dimensional CNN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0376", |
| "paper_id": "2025ApJS..279...46L", |
| "question": "For forecasting major solar flares from magnetograms obtained with different instruments, which ML pipeline is more reliable?", |
| "option_a": "CNN-bidirectional LSTM on magnetogram sequences", |
| "option_b": "inverted Transformer on magnetic-field-feature sequences", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0377", |
| "paper_id": "2025ApJS..280...10C", |
| "question": "For identifying Wolf-Rayet stars from stellar spectra, which ML method gives higher F1 scores?", |
| "option_a": "wide-kernel CNN", |
| "option_b": "wide-kernel CNN with LSTM", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0378", |
| "paper_id": "2025ApJS..280...22Z", |
| "question": "For predicting the solar 10.7 cm radio flux over the next few days, which ML pipeline gives lower error?", |
| "option_a": "Mamba model combining flux history and solar images", |
| "option_b": "Mamba time-series model using flux history", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0379", |
| "paper_id": "2025ApJS..280...22Z", |
| "question": "For predicting the solar 10.7 cm radio flux from recent flux measurements and solar images, which multimodal architecture gives lower error?", |
| "option_a": "Mamba encoders for flux and image sequences", |
| "option_b": "LSTM flux encoder with a Mamba image encoder", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0380", |
| "paper_id": "2025ApJS..281...31Y", |
| "question": "For forecasting daily full-disk solar flare activity from its recent history and related solar observations, which Transformer architecture gives lower mean absolute error?", |
| "option_a": "temporal autocorrelation Transformer", |
| "option_b": "cross-variable attention Transformer", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0381", |
| "paper_id": "2025ApJS..281...42S", |
| "question": "For predicting positions and velocities in three-body simulations, which neural-network architecture gives lower errors?", |
| "option_a": "feed-forward neural network", |
| "option_b": "self-attention neural network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0382", |
| "paper_id": "2025ApJS..281...48X", |
| "question": "For restoring saturated solar extreme-ultraviolet images, which neural-network approach gives more accurate reconstructions?", |
| "option_a": "physics-guided U-Net", |
| "option_b": "generative-adversarial image-restoration network", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0383", |
| "paper_id": "2025MNRAS.544.3104X", |
| "question": "For identifying Ap stars from low-resolution spectra, which method gives a higher F1 score?", |
| "option_a": "channel-attention CNN", |
| "option_b": "XGBoost", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0384", |
| "paper_id": "2025MNRAS.544.3104X", |
| "question": "For identifying Ap stars from low-resolution spectra, which method gives a higher F1 score?", |
| "option_a": "residual CNN with channel attention", |
| "option_b": "residual CNN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0385", |
| "paper_id": "2025PASP..137c4502L", |
| "question": "For classifying astronomical image quality, which CNN architecture gives a higher F1 score?", |
| "option_a": "reduced-block, narrower-channel ShuffleNetV2 with spatial attention", |
| "option_b": "reduced-block, narrower-channel ShuffleNetV2 without spatial attention", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0386", |
| "paper_id": "2025PASP..137f4504Y", |
| "question": "For identifying strong gravitational lenses in galaxy images, which CNN training strategy gives higher recall?", |
| "option_a": "CNN with momentum-contrast self-supervised pretraining", |
| "option_b": "CNN trained only with labeled examples", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0387", |
| "paper_id": "2025PASP..137f4504Y", |
| "question": "For identifying strong gravitational lenses in galaxy images, which architecture gives higher precision?", |
| "option_a": "hybrid CNN-Transformer", |
| "option_b": "Transformer classifier", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0388", |
| "paper_id": "2025PASP..137g4502X", |
| "question": "For reconstructing high-resolution solar H-alpha spectra from low-resolution spectra, which generative model produces less spectral distortion?", |
| "option_a": "wavelet-pooled conditional diffusion model", |
| "option_b": "super-resolution generative adversarial network", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0389", |
| "paper_id": "2025PASP..137i4502C", |
| "question": "For estimating red-giant atmospheric parameters from low-resolution spectra, which CNN architecture gives lower errors?", |
| "option_a": "ConvNeXt", |
| "option_b": "shallow, wide CNN with limited pooling", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0390", |
| "paper_id": "2025PASP..137i4502C", |
| "question": "For estimating red-giant atmospheric parameters from spectra and photometric images, which multimodal network gives lower errors?", |
| "option_a": "multimodal network without a Transformer fusion encoder", |
| "option_b": "multimodal network with a Transformer fusion encoder", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0391", |
| "paper_id": "2025PASP..137j4503L", |
| "question": "For identifying blue straggler stars from low-resolution spectra, which CNN architecture gives a higher F1 score?", |
| "option_a": "ConvNeXt CNN", |
| "option_b": "multiscale dilated large-kernel CNN with grouped channel attention", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0392", |
| "paper_id": "2025PhRvD.112h4025S", |
| "question": "For generating eccentric binary-black-hole waveforms, which neural-network architecture yields lower waveform mismatch?", |
| "option_a": "multilayer perceptron with convolutional refinement", |
| "option_b": "Fourier-feature neural network with convolutional refinement", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0393", |
| "paper_id": "2025RAA....25e5007S", |
| "question": "For identifying pulsar candidates from pulse-profile and dispersion plots, which CNN architecture gives a higher F1 score?", |
| "option_a": "ResNet-50", |
| "option_b": "two-layer CNN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0394", |
| "paper_id": "2025RAA....25i5017Y", |
| "question": "For classifying stars from low-resolution spectra, which deep-learning pipeline gives higher accuracy?", |
| "option_a": "ResNet-DenseNet pipeline", |
| "option_b": "Swin Transformer-ResNet-DenseNet pipeline", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0395", |
| "paper_id": "2025Univ...11...13J", |
| "question": "For identifying Blazhko modulation in RRab stars from their light curves, which ML pipeline gives higher classification accuracy?", |
| "option_a": "residual CNN using phase-folded light curves", |
| "option_b": "multilayer perceptron using light-curve parameters", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0396", |
| "paper_id": "2026MNRAS.547ag445S", |
| "question": "For identifying lensed quasars from multiband images with limited labeled data, which ML pipeline gives a higher F1 score?", |
| "option_a": "variational-autoencoder compression + dense neural-network classifier", |
| "option_b": "convolutional neural network with virtual adversarial training", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0397", |
| "paper_id": "2026PASP..138a4103J", |
| "question": "For forecasting kilonova light curves from gravitational-wave alerts, which neural-network pipeline gives lower prediction error?", |
| "option_a": "bidirectional LSTM using alert parameters", |
| "option_b": "CNN-bidirectional LSTM using alert parameters and sky maps", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0398", |
| "paper_id": "2026A&A...707A.105T", |
| "question": "For estimating photometric redshifts of faint galaxies in fields observed under conditions that differ from the training field, which CNN strategy gives lower scatter and fewer catastrophic errors?", |
| "option_a": "CNN trained only on the labeled field", |
| "option_b": "CNN with adversarial domain adaptation", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0399", |
| "paper_id": "2026A&A...707A.105T", |
| "question": "When target-field spectroscopy is limited mainly to brighter galaxies, which CNN training strategy gives lower bias and fewer catastrophic redshift errors across the full field?", |
| "option_a": "CNN retrained with the available target-field spectra", |
| "option_b": "adversarial domain-adaptation CNN using unlabeled target-field images", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0400", |
| "paper_id": "2026ApJ...998..154K", |
| "question": "For distinguishing repeating and nonrepeating fast radio bursts using their morphology, which ML pipeline gives a higher F1 score?", |
| "option_a": "ConvNeXt using dynamic spectra", |
| "option_b": "gradient-boosted decision trees using fitted burst parameters", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0401", |
| "paper_id": "2026ApJ...998..340L", |
| "question": "For predicting whether a solar flare will be accompanied by a coronal mass ejection from active-region magnetograms, which neural-network pipeline gives a higher TSS?", |
| "option_a": "vision Transformer and LSTM using magnetogram sequences", |
| "option_b": "vision Transformer using individual magnetograms", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0402", |
| "paper_id": "2026A&A...706A.317C", |
| "question": "For locating brightest cluster galaxies in simulated multiband images, which residual-network architecture gives more accurate positions?", |
| "option_a": "residual encoder-decoder network predicting a position map", |
| "option_b": "residual network directly regressing position coordinates", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0403", |
| "paper_id": "2026ApJ...996....9O", |
| "question": "For modeling the thermal and chemical evolution of primordial gas with many short timesteps, which neural-network pipeline remains accurate?", |
| "option_a": "deep operator network with learned timescale-guided updates", |
| "option_b": "deep operator network with direct iterative updates", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0404", |
| "paper_id": "2026A&A...705A.226B", |
| "question": "For separating emission components in X-ray images when the component spectra differ from those represented in the training data, which ML pipeline is more robust?", |
| "option_a": "unrolled proximal-optimization network", |
| "option_b": "iterative optimization with an embedded learned wavelet-like CNN transform", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0405", |
| "paper_id": "2026A&A...708A..23A", |
| "question": "For identifying hot-subdwarf binaries with composite spectra, which ML method gives more accurate classifications?", |
| "option_a": "self-organizing map classifier", |
| "option_b": "ensemble of convolutional neural networks", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0406", |
| "paper_id": "2026A&A...708A..23A", |
| "question": "For distinguishing cool, helium-poor hot subdwarfs from hotter, helium-rich hot subdwarfs in low-resolution spectra, which ML method gives more accurate classifications?", |
| "option_a": "ensemble of convolutional neural networks", |
| "option_b": "self-organizing map classifier", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0407", |
| "paper_id": "2026ApJ..1002L..47K", |
| "question": "For recovering the Galactic mass distribution from stellar kinematics in low-completeness regions, which normalizing-flow pipeline is more accurate?", |
| "option_a": "conditional velocity-distribution and stellar-density modeling", |
| "option_b": "selection-weighted phase-space modeling", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0408", |
| "paper_id": "2026A&A...709A..90C", |
| "question": "For estimating sodium abundances of late-type stars from a three-dimensional non-LTE spectral grid, which ML pipeline gives lower overall interpolation error?", |
| "option_a": "radial-basis-function interpolation of line profiles", |
| "option_b": "feed-forward neural network using line strengths", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0409", |
| "paper_id": "2026ApJ..1000...14F", |
| "question": "For classifying supernova subtypes from low-resolution spectra, which neural-network pipeline gives a higher macro F1 score?", |
| "option_a": "convolutional spectral classifier", |
| "option_b": "masked-spectrum-pretrained attention classifier", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0410", |
| "paper_id": "2026ApJ..1004..168J", |
| "question": "For reconstructing the three-dimensional evolution of a coronal mass ejection from simulated coronagraph observations with few viewing directions, which neural-network variant gives lower electron-density error?", |
| "option_a": "neural radiance field with plasma-continuity and solar-wind constraints", |
| "option_b": "neural radiance field without physical constraints", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0411", |
| "paper_id": "2026ApJS..283...45C", |
| "question": "For detecting fast radio bursts in dynamic spectra, which neural-network architecture gives a higher F1 score?", |
| "option_a": "residual CNN", |
| "option_b": "Swin Transformer", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0412", |
| "paper_id": "2026ApJS..283...45C", |
| "question": "For estimating the dispersion measure of fast radio bursts, which ML approach gives more precise estimates?", |
| "option_a": "direct neural-network regression", |
| "option_b": "neural segmentation followed by dispersion-curve fitting", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0413", |
| "paper_id": "2026A&A...708A.280B", |
| "question": "For inferring spectral-model parameters from simulated high-resolution X-ray spectra, which dimensionality-reduction method gives more accurate posterior distributions?", |
| "option_a": "principal component analysis", |
| "option_b": "autoencoder", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0414", |
| "paper_id": "2026ApJ...998..102W", |
| "question": "For identifying quasi-periodic micropulses in individual pulsar pulses, which residual-network pipeline gives higher precision at similar recall?", |
| "option_a": "two-stage residual network combining pulse profiles and power-spectrum diagnostics", |
| "option_b": "single-stage residual network using power-spectrum diagnostics", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0415", |
| "paper_id": "2026ApJ...998..102W", |
| "question": "For identifying quasi-periodic micropulses in individual pulsar pulses, which residual-network pipeline gives better classification performance?", |
| "option_a": "two-stage residual network using pulse-profile and power-spectrum images", |
| "option_b": "single residual network combining one-dimensional pulse and power-spectrum sequences", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0416", |
| "paper_id": "2026ApJS..283...55B", |
| "question": "For classifying supernovae from host-galaxy information, which ML pipeline gives a higher class-weighted F1 score?", |
| "option_a": "random forest using neural-network estimates of galaxy stellar mass and star-formation rate, redshift, and host separation", |
| "option_b": "random forest using host-galaxy photometry", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0417", |
| "paper_id": "2026ApJ..1001..120X", |
| "question": "For estimating the position angles of protostellar outflows from simulated CO observations when the training set does not cover every mass and inclination, which architecture gives lower prediction error?", |
| "option_a": "multimodal model with a Vision Transformer image encoder", |
| "option_b": "multimodal model with a residual CNN image encoder", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0418", |
| "paper_id": "2026ApJ..1003..141D", |
| "question": "For classifying variable stars from long, densely sampled light curves, which sequence architecture gives higher macro-F1?", |
| "option_a": "Transformer", |
| "option_b": "LSTM", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0419", |
| "paper_id": "2026ApJ..1003..141D", |
| "question": "For classifying variable stars from light curves, which contrastive-learning strategy gives higher macro-F1?", |
| "option_a": "non-hierarchical contrastive learning", |
| "option_b": "hierarchical contrastive learning", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0420", |
| "paper_id": "2026A&A...707A.203C", |
| "question": "For detecting H I galaxies in simulated radio data cubes when accurate source localization and sizing are required, which 3D CNN variant performs better?", |
| "option_a": "3D CNN with a cascaded loss for joint detection and source-parameter prediction", |
| "option_b": "3D CNN with a detection-only loss", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0421", |
| "paper_id": "2026ApJS..283....7S", |
| "question": "For distinguishing eruptive and confined solar flares using light-curve morphology measurements and CNN-derived features, which classifier better separates the two classes while producing fewer false alarms?", |
| "option_a": "random forest", |
| "option_b": "linear support-vector machine", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0422", |
| "paper_id": "2026A&A...706A.157P", |
| "question": "For interpolating three-dimensional temperature and horizontal wind fields across a grid of simulated gas-giant atmospheres, which ML model gives lower prediction error?", |
| "option_a": "XGBoost", |
| "option_b": "dense neural network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0423", |
| "paper_id": "2026ApJS..282...59Z", |
| "question": "For distinguishing gamma-ray bursts from background in multi-energy-band light curves, which residual CNN achieves higher precision at similar recall?", |
| "option_a": "residual CNN with adaptive frequency-domain enhancement", |
| "option_b": "residual CNN with channel-and-spatial attention", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0424", |
| "paper_id": "2026PASP..138e4508J", |
| "question": "For classifying transients and variable sources from early multiband light curves, which Transformer pipeline gives higher accuracy?", |
| "option_a": "sparse-attention Transformer", |
| "option_b": "dense-attention Transformer with masked-light-curve pretraining", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0425", |
| "paper_id": "2026PASP..138e4508J", |
| "question": "For classifying transients and variable sources from spectra, which CNN architecture gives higher accuracy?", |
| "option_a": "multiscale 1D CNN with channel attention", |
| "option_b": "conventional 1D CNN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0426", |
| "paper_id": "2026PhRvD.113d3522C", |
| "question": "For estimating the amplitude of the stochastic gravitational-wave background from simulated quasar proper-motion catalogs with complete sky coverage, which neural-network architecture gives lower prediction error?", |
| "option_a": "fully connected neural network", |
| "option_b": "graph neural network", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0427", |
| "paper_id": "2026PhRvD.113j3536Z", |
| "question": "Which ML pipeline more accurately predicts cosmic microwave background temperature and polarization power spectra from cosmological parameters?", |
| "option_a": "residual multilayer perceptron predicting PCA coefficients and reconstructing the spectra", |
| "option_b": "self-attention Transformer directly predicting the power spectra", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0428", |
| "paper_id": "2026PhRvD.113j3536Z", |
| "question": "For predicting cosmic microwave background temperature and polarization power spectra with an attention-based emulator, which attention mechanism gives higher accuracy?", |
| "option_a": "dot-product attention", |
| "option_b": "linear attention", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0429", |
| "paper_id": "2026A&C....54.01027S", |
| "question": "For estimating the parameters of non-precessing binary black holes from simulated gravitational-wave signals, which neural-network architecture gives lower prediction error?", |
| "option_a": "CNN", |
| "option_b": "hybrid CNN-Transformer", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0430", |
| "paper_id": "2026ApJS..282...25M", |
| "question": "For identifying early-type galaxies with dust lanes from galaxy images, which ML pipeline gives a higher F1 score?", |
| "option_a": "focal-loss CNN", |
| "option_b": "CNN trained with conditional-diffusion augmentation", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0431", |
| "paper_id": "2026AJ....171..268F", |
| "question": "For generating galaxy images in specified morphological classes, which diffusion-model architecture produces more realistic images?", |
| "option_a": "conditional diffusion model with additional U-Net skip connections", |
| "option_b": "conditional diffusion model with a standard U-Net", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0432", |
| "paper_id": "2026AJ....171..268F", |
| "question": "For generating galaxy images, which diffusion-model architecture more accurately reproduces their color and size distributions?", |
| "option_a": "U-Net diffusion model with adaptive convolution-attention blocks", |
| "option_b": "U-Net diffusion model without adaptive convolution-attention blocks", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0433", |
| "paper_id": "2026MNRAS.546ag283T", |
| "question": "For predicting the infrared spectra of polycyclic aromatic hydrocarbons from their molecular structures, which method gives lower spectral error?", |
| "option_a": "graph neural network with attention-based molecular aggregation", |
| "option_b": "graph attention network", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0434", |
| "paper_id": "2026ApJ...999..231S", |
| "question": "For localizing gamma-ray bursts in simulated observations with low photon counts and strong background contamination, which neural-network architecture gives smaller angular offsets?", |
| "option_a": "reconstructed-image U-Net", |
| "option_b": "hybrid detector-event and reconstructed-image network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0435", |
| "paper_id": "2026ApJ...999..231S", |
| "question": "For localizing gamma-ray bursts in simulated observations with low photon counts and strong background contamination, which neural-network architecture gives smaller angular offsets?", |
| "option_a": "detector-event MLP reconstruction network", |
| "option_b": "hybrid detector-event and reconstructed-image network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0436", |
| "paper_id": "2026ApJ...997...36M", |
| "question": "For estimating photometric redshifts among galaxies with similar colors, which ML approach is less biased toward the mean redshift of the group?", |
| "option_a": "image-based deep capsule network", |
| "option_b": "local linear regression using integrated photometry", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0437", |
| "paper_id": "2026PhRvD.113j3548W", |
| "question": "For recovering the cosmological 21-cm signal from simulated foreground-contaminated intensity maps, which ML pipeline is more accurate?", |
| "option_a": "PCA foreground subtraction followed by U-Net reconstruction", |
| "option_b": "PCA foreground subtraction", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0438", |
| "paper_id": "2026PhRvD.113d3508Y", |
| "question": "For predicting nonlinear matter power spectra from cosmological parameters using simulations of different fidelity, which neural-network architecture gives lower mean error?", |
| "option_a": "low-fidelity emulator with a correction network directly predicting the high-fidelity spectrum", |
| "option_b": "low-fidelity emulator with a correction network predicting the high-to-low-fidelity ratio", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0439", |
| "paper_id": "2026ApJ..1002..142D", |
| "question": "For estimating the three-dimensional shapes of massive galaxy clusters from simulated observations, which neural-network pipeline is more accurate?", |
| "option_a": "CNN-GNN fusion network using cluster images and member-galaxy measurements", |
| "option_b": "GNN using member-galaxy measurements", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0440", |
| "paper_id": "2026ApJS..284...58Z", |
| "question": "For identifying observations containing fast radio transients from time-frequency spectra, which instance-segmentation backbone produces fewer false detections while retaining all transient-containing observations?", |
| "option_a": "ResNet-18 residual CNN backbone", |
| "option_b": "HRNet high-resolution CNN backbone", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0441", |
| "paper_id": "2026AJ....171..114S", |
| "question": "For estimating galaxy redshifts from broadband photometry using a combined set of spectroscopic and photometric reference redshifts, which neural-network architecture gives lower redshift-estimation scatter?", |
| "option_a": "Bayesian neural network", |
| "option_b": "point-estimate neural network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0442", |
| "paper_id": "2026AJ....171..114S", |
| "question": "For estimating galaxy redshifts from broadband photometry using a combined set of spectroscopic and photometric reference redshifts, which neural-network approach provides more reliable calibrated redshift uncertainties?", |
| "option_a": "point-estimate neural network", |
| "option_b": "Bayesian neural network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0443", |
| "paper_id": "2026ApJS..283...30S", |
| "question": "For classifying transient reports by observational band or messenger, which text-embedding approach gives higher accuracy?", |
| "option_a": "pretrained Sentence Transformer", |
| "option_b": "contrastively fine-tuned Sentence Transformer", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0444", |
| "paper_id": "2026ApJ..1002..215B", |
| "question": "For predicting the long-term ejection of near-Earth asteroids from their initial orbital elements, which supervised classifier gives higher accuracy?", |
| "option_a": "k-nearest-neighbor classifier", |
| "option_b": "gradient-boosted decision trees", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0445", |
| "paper_id": "2026A&A...710A.194B", |
| "question": "For removing stellar-activity signals from simulated radial-velocity measurements using changes in spectral line profiles, which neural-network architecture leaves lower residual variability?", |
| "option_a": "fully connected neural network", |
| "option_b": "convolutional attention network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0446", |
| "paper_id": "2026A&A...710A.194B", |
| "question": "For removing stellar-activity signals from radial-velocity observations of an active K dwarf, which neural-network pipeline leaves lower residual variability?", |
| "option_a": "convolutional network using standard line-profile indicators", |
| "option_b": "convolutional attention network using orthogonal line-profile coefficients", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0447", |
| "paper_id": "2026A&C....5401002K", |
| "question": "For detecting simulated fast radio bursts from few-channel radio time series, which CNN architecture gives higher recall?", |
| "option_a": "one-dimensional fully convolutional network", |
| "option_b": "one-dimensional EfficientNet CNN", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0448", |
| "paper_id": "2026A&C....5401002K", |
| "question": "For searching few-channel radio time series for fast radio bursts, which CNN pipeline produces fewer false detections?", |
| "option_a": "single EfficientNet classifier", |
| "option_b": "two-stage EfficientNet cascade with the second classifier trained on first-stage false positives", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0449", |
| "paper_id": "2026ApJ..1003...27S", |
| "question": "For predicting stellar spectra from stellar parameters and chemical abundances, which activation function gives lower validation error?", |
| "option_a": "rectified linear unit", |
| "option_b": "sigmoid linear unit", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0450", |
| "paper_id": "2026AJ....171..340K", |
| "question": "For estimating galaxy redshifts from deep, high-resolution images, which CNN training strategy gives more accurate predictions?", |
| "option_a": "contrastive pretraining on galaxy images followed by redshift fine-tuning", |
| "option_b": "joint contrastive, color-prediction, and redshift-prediction training", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0451", |
| "paper_id": "2026ApJ...997...65R", |
| "question": "For estimating galaxy redshifts from multiband photometry, which normalizing-flow architecture produces fewer catastrophic errors?", |
| "option_a": "conditional affine-coupling normalizing flow", |
| "option_b": "conditional neural spline flow", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0452", |
| "paper_id": "2026ApJ...997...65R", |
| "question": "For estimating galaxy redshifts from broadband photometry, which generative model gives more accurate estimates?", |
| "option_a": "conditional generative adversarial network", |
| "option_b": "conditional normalizing flow", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0453", |
| "paper_id": "2026ApJ...996..107B", |
| "question": "For estimating galaxy redshifts from broadband photometry, which ML approach gives more accurate estimates?", |
| "option_a": "weighted ensemble of gradient-boosted trees, nearest-neighbor regression, and a neural network", |
| "option_b": "gradient-boosted decision-tree regression", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0454", |
| "paper_id": "2026PASP..138c4503R", |
| "question": "For classifying transient candidates from image cutouts, which training strategy gives higher ROC AUC?", |
| "option_a": "MaxViT pretrained on galaxy images", |
| "option_b": "MaxViT without pretraining", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0455", |
| "paper_id": "2026PASP..138c4503R", |
| "question": "For real-time classification of transient candidates from image cutouts, which vision architecture is more computationally efficient?", |
| "option_a": "ConvNeXt CNN", |
| "option_b": "MaxViT", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0456", |
| "paper_id": "2026PASP..138f4501L", |
| "question": "For classifying fields by stellar density from simulated multiband images, which CNN architecture provides higher accuracy with lower computational cost?", |
| "option_a": "DenseNet-121", |
| "option_b": "ResNet-34", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0457", |
| "paper_id": "2026A&C....5501067S", |
| "question": "For estimating cosmological parameters from simulated X-ray cluster catalogs, which neural-network pipeline gives lower prediction error?", |
| "option_a": "fully connected network using cluster abundance and power spectra", |
| "option_b": "3D CNN using cluster abundance and spatial density fields", |
| "correct_option": "B", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0458", |
| "paper_id": "2026AJ....171..205I", |
| "question": "For distinguishing transients from image artifacts using a large labeled training set, which neural-network architecture gives higher classification accuracy?", |
| "option_a": "convolutional neural network", |
| "option_b": "localized-attention network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0459", |
| "paper_id": "2026A&C....5400997M", |
| "question": "For solving stiff Lane-Emden equations, which neural architecture gives lower solution error?", |
| "option_a": "fully connected physics-informed network", |
| "option_b": "physics-informed network combining radial-basis-function and Kolmogorov-Arnold layers", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0460", |
| "paper_id": "2026A&C....5501037C", |
| "question": "For distinguishing primary particle-track hits from backscattered hits in strongly imbalanced simulated tracker data, which training objective gives a higher F1 score?", |
| "option_a": "graph-balanced focal loss", |
| "option_b": "binary cross-entropy loss", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0461", |
| "paper_id": "2026A&C....5501037C", |
| "question": "For distinguishing primary particle-track hits from backscattered hits in simulated tracker data, which interaction-network architecture gives higher classification accuracy?", |
| "option_a": "interaction network with a multi-head attention encoder", |
| "option_b": "interaction network with a fully connected encoder", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0462", |
| "paper_id": "2026AJ....171...94C", |
| "question": "For classifying radio-frequency interference in pulsar frequency-phase images, which image-classification architecture gives a higher F1 score?", |
| "option_a": "MobileNetV2 with channel attention, dynamic activations, and multiscale feature fusion", |
| "option_b": "Swin Transformer", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0463", |
| "paper_id": "2026AJ....171..107C", |
| "question": "For estimating spatially structured backgrounds in astronomical images, which generative model gives lower reconstruction error?", |
| "option_a": "denoising diffusion probabilistic model", |
| "option_b": "generative adversarial inpainting network", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0464", |
| "paper_id": "2026AJ....171..132C", |
| "question": "For classifying galaxy morphologies, which training strategy gives more accurate classifications?", |
| "option_a": "Vision Transformer with vision-language contrastive pretraining", |
| "option_b": "Vision Transformer with image-classification pretraining", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0465", |
| "paper_id": "2026AJ....171..311W", |
| "question": "For estimating stellar parameters from spectra, which neural-network pipeline gives lower prediction error?", |
| "option_a": "gated recurrent neural network using spectra", |
| "option_b": "Transformer combining spectra and photometry", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0466", |
| "paper_id": "2026AJ....172...18M", |
| "question": "For reconstructing asteroid gravitational fields from simulated measurements, which neural-network pipeline gives lower relative error?", |
| "option_a": "neural density-field model with shape information, gravitational-equation regularization, and adaptive altitude sampling", |
| "option_b": "neural density-field model with uniform spatial sampling", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0467", |
| "paper_id": "2026ApJ...997..135P", |
| "question": "For localizing short gamma-ray bursts from detector-panel count rates in background-free simulations, which neural-network output representation gives lower angular error?", |
| "option_a": "sine-cosine angle regression", |
| "option_b": "direct angle regression", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0468", |
| "paper_id": "2026ApJ...998..258Y", |
| "question": "For identifying severely inaccurate galaxy photometric-redshift estimates, which ML pipeline gives higher recall with fewer false positives?", |
| "option_a": "binary neural network using galaxy magnitudes directly", |
| "option_b": "binary neural network using redshift probabilities generated by a support-vector machine", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0469", |
| "paper_id": "2026ApJ..1001...51Z", |
| "question": "For inferring the parameters of a nuclear level-density formula, which Transformer training strategy gives lower prediction error?", |
| "option_a": "Transformer trained without physical-constraint penalties", |
| "option_b": "Transformer trained with physical-constraint penalties", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0470", |
| "paper_id": "2026ApJ..1001...51Z", |
| "question": "For predicting nuclear level-density parameters for an isotope chain excluded from training, which Transformer training strategy gives lower error?", |
| "option_a": "Transformer trained with physical constraints and no parameter priors", |
| "option_b": "Transformer initialized from parameter priors without physical constraints", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0471", |
| "paper_id": "2026ApJ..1001...96C", |
| "question": "For extrapolating coronal magnetic fields, which generative-model training strategy produces fields that better satisfy the force-free and divergence-free conditions?", |
| "option_a": "3D generative adversarial network with equation-constrained retraining and numerical-gradient guidance", |
| "option_b": "3D generative adversarial network without equation-constrained retraining or numerical-gradient guidance", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0472", |
| "paper_id": "2026ApJS..282....5Z", |
| "question": "For reconstructing irregular gaps in neutron-monitor count rates, which method gives lower error?", |
| "option_a": "TimesNet periodicity-based convolutional network", |
| "option_b": "SAITS self-attention imputation network", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0473", |
| "paper_id": "2026ApJS..282...52L", |
| "question": "For multi-day forecasting of the 10.7 cm solar radio flux from its recent history, which neural-network architecture gives lower prediction error?", |
| "option_a": "TimesNet with frequency-domain convolution", |
| "option_b": "N-BEATS with residual fully connected blocks", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0474", |
| "paper_id": "2026ApJS..282...70C", |
| "question": "For classifying chemically peculiar stars from low-resolution spectra, which ML method gives more accurate classifications?", |
| "option_a": "hybrid LSTM-CNN", |
| "option_b": "XGBoost", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0475", |
| "paper_id": "2026ApJS..283...38S", |
| "question": "For identifying quasars from low-resolution spectra and catalog measurements, which deep-learning pipeline gives a higher F1 score?", |
| "option_a": "neural network combining spectra and catalog measurements", |
| "option_b": "Transformer using catalog measurements only", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0476", |
| "paper_id": "2026ApJS..283...44L", |
| "question": "For classifying solar active regions by magnetic type from magnetograms and continuum images, which neural-network architecture gives higher overall accuracy?", |
| "option_a": "vision Transformer", |
| "option_b": "hybrid vision Transformer-CNN", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0477", |
| "paper_id": "2026ApJS..283...46L", |
| "question": "For predicting the maximum X-ray flux from solar active-region magnetogram sequences over the next day, which neural-network method gives lower error?", |
| "option_a": "CNN-LSTM with standard point-estimate output", |
| "option_b": "CNN-LSTM with Monte Carlo-dropout probabilistic output", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0478", |
| "paper_id": "2026ApJS..283...49L", |
| "question": "For classifying galaxy morphologies from images, which vision Transformer training strategy gives higher accuracy?", |
| "option_a": "vision Transformer trained from scratch", |
| "option_b": "vision Transformer with masked-autoencoder pretraining", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0479", |
| "paper_id": "2026ApJS..283...49L", |
| "question": "For detecting strong gravitational lenses in images, which vision Transformer training strategy gives higher average precision?", |
| "option_a": "vision Transformer trained from scratch", |
| "option_b": "vision Transformer with masked-autoencoder pretraining", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0480", |
| "paper_id": "2026ApJS..283...56Y", |
| "question": "For extrapolating coronal magnetic fields from photospheric magnetograms, which U-shaped neural-operator training strategy better satisfies the force-free condition?", |
| "option_a": "U-shaped neural operator trained only on magnetic-field data", |
| "option_b": "U-shaped neural operator fine-tuned with grid-based physical constraints", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0481", |
| "paper_id": "2026ApJS..283...58L", |
| "question": "For forecasting M-class or stronger solar flares from active-region data, which ML pipeline gives a higher F1 score?", |
| "option_a": "fusion of a solar-activity-parameter network and a magnetogram CNN", |
| "option_b": "magnetogram CNN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0482", |
| "paper_id": "2026ApJS..284....7L", |
| "question": "For estimating elemental abundances by fitting simulated stellar spectra from the training distribution, which parameter-fitting strategy gives lower errors?", |
| "option_a": "fitting all PCA combinations of stellar parameters with L1 regularization", |
| "option_b": "fitting only the dominant PCA combinations of stellar parameters", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0483", |
| "paper_id": "2026ApJS..284....7L", |
| "question": "For estimating stellar parameters by fitting observed spectra with a neural spectral model built from simulated spectra, which strategy agrees more closely with catalog values?", |
| "option_a": "fitting only the dominant PCA combinations of stellar parameters", |
| "option_b": "fitting all PCA combinations of stellar parameters with L1 regularization", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0484", |
| "paper_id": "2026ApJS..284...14L", |
| "question": "For identifying Wolf-Rayet stars from stellar spectra, which ML method gives a higher F1 score?", |
| "option_a": "random forest using the same emission-line-guided features", |
| "option_b": "multiscale large-kernel CNN with emission-line-guided attention", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0485", |
| "paper_id": "2026ApJS..284...14L", |
| "question": "For distinguishing Wolf-Rayet subtypes from other stellar spectra, which CNN variant gives a higher F1 score?", |
| "option_a": "the same CNN with random initialization", |
| "option_b": "multiscale large-kernel CNN initialized with broad emission-line profiles", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0486", |
| "paper_id": "2026ApJS..284...27M", |
| "question": "For localizing galaxy-scale strong lenses when accurate bounding boxes are required, which transformer detector gives higher average precision?", |
| "option_a": "the same transformer detector with auxiliary arc supervision only", |
| "option_b": "transformer detector with auxiliary arc supervision and explicit lens-arc relation modeling", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0487", |
| "paper_id": "2026ApJS..284...35L", |
| "question": "For reconstructing cosmic microwave background B modes from simulated single-frequency polarization maps contaminated by Galactic dust, which neural-network pipeline gives lower reconstruction error?", |
| "option_a": "self-attention adversarial map-reconstruction network", |
| "option_b": "U-Net regression map-reconstruction network", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0488", |
| "paper_id": "2026ApJS..284...45W", |
| "question": "For estimating stellar masses of low-redshift galaxies from photometric survey data, which neural-network pipeline gives lower prediction error?", |
| "option_a": "ResNet using galaxy images", |
| "option_b": "ResNet-MLP combining galaxy images and catalog photometry", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0489", |
| "paper_id": "2026ApJS..284...66L", |
| "question": "For predicting gamma-ray detector background spectra from spacecraft telemetry, which neural-network architecture gives lower error?", |
| "option_a": "grouped Transformer with single-layer perceptron encoders", |
| "option_b": "grouped Transformer with three-layer MLP encoders", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0490", |
| "paper_id": "2026ApJS..284...74H", |
| "question": "For detecting coronal mass ejections in coronagraph images, which self-supervised image encoder gives a higher F1 score?", |
| "option_a": "LeNet encoder trained by masked image reconstruction", |
| "option_b": "vision Transformer encoder trained by masked image reconstruction", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0491", |
| "paper_id": "2026ApJS..284...74H", |
| "question": "For localizing coronal mass ejections in coronagraph images, which PCA-based feature extractor gives a higher F1 score?", |
| "option_a": "two-stage PCA convolutional network", |
| "option_b": "single-layer PCA projection", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0492", |
| "paper_id": "2026ApJS..284...78Z", |
| "question": "For segmenting radio-frequency interference in radio-telescope time-frequency data when only a small fraction of the samples are labeled, which training strategy gives higher F1 and boundary accuracy?", |
| "option_a": "adaptive-threshold semisupervised training with two teachers and boundary-aware loss", |
| "option_b": "supervised training using labeled samples only", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0493", |
| "paper_id": "2026ApJS..284...78Z", |
| "question": "For segmenting radio-frequency interference in radio-telescope time-frequency data with very few labeled examples, which CNN architecture gives higher F1 and boundary accuracy?", |
| "option_a": "multiscale attention residual U-Net", |
| "option_b": "lightweight residual fully convolutional network", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0494", |
| "paper_id": "2026MNRAS.545f2095T", |
| "question": "For estimating black-hole spin from simulated images at the angular resolution of horizon-scale observations, which neural-network pipeline gives more accurate predictions?", |
| "option_a": "autoencoder super-resolution followed by CNN regression", |
| "option_b": "CNN-RNN regression from image sequences", |
| "correct_option": "A", |
| "difficulty": "hard" |
| }, |
| { |
| "question_id": "Q0495", |
| "paper_id": "2026PASP..138d4508H", |
| "question": "For classifying galaxy morphologies when the test images come from a different survey than the training images, which ML approach is more accurate?", |
| "option_a": "pretrained multimodal vision-language model fine-tuned for morphology classification", |
| "option_b": "Swin Transformer trained for morphology classification", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0496", |
| "paper_id": "2026PASP..138e4501L", |
| "question": "For detecting asteroids with faint trails in ground-based images, which object-detection architecture gives higher overall accuracy?", |
| "option_a": "standard real-time detection Transformer", |
| "option_b": "real-time detection Transformer with adaptive multiscale features, dual-path queries, and scale-balanced training", |
| "correct_option": "B", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0497", |
| "paper_id": "2026PASP..138e4503W", |
| "question": "For identifying Galactic H II regions from medium-resolution spectra with limited labeled examples, which neural-network training strategy gives higher classification accuracy?", |
| "option_a": "Siamese metric learning with shared wide-kernel CNN encoders", |
| "option_b": "direct classification with the same wide-kernel CNN", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0498", |
| "paper_id": "2026PASP..138e4505G", |
| "question": "For selecting high-redshift Lyα-blob candidates from broadband images using measured morphology and photometry, which supervised classifier gives a higher F1 score with limited labeled data?", |
| "option_a": "XGBoost", |
| "option_b": "multilayer perceptron", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0499", |
| "paper_id": "2026PASP..138e4509D", |
| "question": "For estimating quasar photometric redshifts from multiband images, which neural-network pipeline gives more accurate predictions?", |
| "option_a": "ResNet with image-spectrum contrastive pretraining", |
| "option_b": "ResNet trained directly for redshift regression from images", |
| "correct_option": "A", |
| "difficulty": "easy" |
| }, |
| { |
| "question_id": "Q0500", |
| "paper_id": "2026Univ...12...50L", |
| "question": "For predicting the intensity of strong sporadic E layers from ionosonde observations, which neural-network design gives lower error?", |
| "option_a": "two-stage network that first predicts occurrence and then estimates event intensity", |
| "option_b": "single-stage network that directly estimates intensity", |
| "correct_option": "A", |
| "difficulty": "easy" |
| } |
| ] |
|
|
