VIP20
本课程属于：029-AI for CAE Simulation VVUQ & Geometric Deep Learning English Technical Courses · 共 20 门相关课程，组合学习更完整。
查看所属VIP全部课程 →
← 返回所属目录 咨询 / 购买

Active Learning and Trusted AI Simulation 30 Cases

30 Chapters
View Course Samples Online
01
Uncertainty Sampling with Margin Strategy on a Synthetic Binary Classifier
Implement a margin-based uncertainty sampling loop on a synthetic two-class dataset and compare label efficiency against random sampling under a fixed labeling budget.
UncertaintySampling
02
Entropy-Based Active Learning for Multiclass Image Feature Classification
Apply entropy uncertainty sampling to a multiclass feature dataset and measure the reduction in labels required to reach a target accuracy versus passive learning.
Entropy-BasedMulticlass
03
Query-by-Committee Disagreement on a Small Ensemble
Build a small ensemble of diverse classifiers, compute committee disagreement, and select the most disputed samples to demonstrate improved decision boundary refinement.
Query-by-CommitteeDisagreement
04
Expected Model Change Selection with a Linear Model
Implement expected model change acquisition for a linear classifier and compare its label efficiency with uncertainty sampling on a controlled dataset.
ExpectedSelection
05
Diversity-Aware Batch Active Learning with k-Means Clustering
Design a batch acquisition function that combines uncertainty with k-means diversity to avoid redundant queries and evaluate batch diversity and accuracy gains.
Diversity-AwareClustering
06
Core-Set Selection for Active Learning
Implement core-set selection using distance to the current labeled set and assess its coverage and label efficiency on a synthetic geometric dataset.
Core-SetSelection
07
Cold-Start Active Learning with Clustering-Based Initialization
Compare random initialization with clustering-based initialization for the first labeled batch and quantify the impact on early learning performance.
Cold-StartClustering-Based
08
Active Learning with Noisy Oracles and Label Correction
Simulate a noisy labeling oracle, implement a simple label correction or re-query strategy, and measure robustness of active learning under label noise.
CorrectionSimulate
09
Cost-Sensitive Active Learning with Variable Labeling Costs
Define per-sample labeling costs and implement a cost-sensitive acquisition function that maximizes expected performance gain per unit cost.
Cost-SensitiveVariable
10
Stream-Based Active Learning with a Budget Constraint
Implement a stream-based selective sampling policy with a fixed budget and evaluate its ability to maintain accuracy while limiting label requests.
Stream-BasedConstraint
11
Pool-Based Active Learning for Regression with Query by Committee
Adapt query-by-committee to a regression task, select samples with high committee variance, and compare label efficiency against random sampling.
Pool-BasedRegression
12
Active Learning for Anomaly Detection with Uncertainty and Density
Combine uncertainty and density-based acquisition to actively label anomalies in an imbalanced dataset and evaluate detection performance with limited labels.
DetectionUncertainty
13
Active Learning with a Semi-Supervised Pseudo-Labeling Loop
Integrate pseudo-labeling of high-confidence unlabeled data into an active learning loop and measure the combined effect on label efficiency.
Semi-SupervisedPseudo-Labeling
14
Deep Active Learning with Monte Carlo Dropout on MNIST
Implement Monte Carlo dropout uncertainty estimation for a small CNN on MNIST and use it for active learning, comparing accuracy per label against random selection.
MNISTImplement
15
Deep Active Learning with Ensemble Uncertainty on CIFAR-10 Subset
Train a small ensemble of CNNs on a CIFAR-10 subset, use ensemble disagreement for acquisition, and analyze the trade-off between ensemble size and label efficiency.
EnsembleUncertainty
16
Active Learning for Text Classification with TF-IDF and Uncertainty Sampling
Apply uncertainty sampling to a TF-IDF text classification pipeline and demonstrate label savings on a small text corpus.
ClassificationTF-IDF
17
Active Learning with Sentence Embeddings and Diversity Sampling
Use sentence embeddings and diversity-based acquisition to select informative text samples and compare with uncertainty sampling on a classification task.
SentenceEmbeddings
18
Active Learning for Named Entity Recognition with Token-Level Uncertainty
Implement token-level uncertainty aggregation for a simple NER model and evaluate active learning label efficiency at the sequence level.
RecognitionToken-Level
19
Active Learning for Object Detection with Localization Uncertainty
Design an acquisition function that combines classification uncertainty and localization uncertainty for object detection and test it on a small detection dataset.
DetectionLocalization
20
Active Learning for Semantic Segmentation with Region Impurity
Compute region impurity from predicted segmentation maps and use it to select images for labeling, measuring improvement in segmentation quality per label.
SemanticSegmentation
21
Active Learning for Time Series Classification with Window Uncertainty
Apply uncertainty sampling at the window level for time series classification and evaluate label efficiency on a sensor dataset.
ClassificationUncertainty
22
Active Learning for Tabular Data with Missing Values
Handle missing values in tabular data and implement an acquisition function that accounts for missingness patterns to improve active learning performance.
implementacquisition
23
Active Learning with a Human-in-the-Loop Simulation Interface
Simulate a human-in-the-loop labeling interface with response delays and implement an asynchronous active learning loop that adapts to delayed feedback.
Human-in-the-LoopSimulation
24
Active Learning under Distribution Shift with Importance Weighting
Introduce a distribution shift between pool and test data, apply importance weighting to acquisition scores, and measure robustness of active learning.
DistributionImportance
25
Active Learning for Fairness-Aware Model Training
Incorporate fairness constraints into acquisition by prioritizing samples from underrepresented groups and evaluate fairness metrics alongside accuracy.
Fairness-AwareTraining
26
Active Learning with Explainability-Guided Acquisition
Use model explanations (e.g., SHAP values) to identify uncertain or influential samples for labeling and compare with standard uncertainty sampling.
Explainability-GuidedAcquisition
27
Active Learning for Trusted AI: Robustness Testing with Adversarial Queries
Generate adversarial examples as candidate queries and design an acquisition strategy that improves model robustness with minimal labels.
RobustnessAdversarial
28
Active Learning for Calibration: Selecting Samples to Improve Confidence Calibration
Implement an acquisition function that targets miscalibrated regions and evaluate expected calibration error improvement per label.
CalibrationSelecting
29
Active Learning Simulation for Trustworthy Model Auditing
Simulate an auditing scenario where active learning is used to efficiently uncover model failures, and report detection rate and label cost.
SimulationTrustworthy
30
End-to-End Active Learning Pipeline with Experiment Tracking and Reproducibility
Assemble a complete active learning pipeline with configuration management, experiment tracking, and reproducibility checks, and compare multiple acquisition strategies in a single run.
End-to-EndPipeline