VIP20
本课程属于：005-Artificial Intelligence and Machine Learning Engineering 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

Robust Machine Learning Systems Under Data Drift and Noise

30 Chapters
View Course Samples Online
01
Course overview
scope, concepts, and technical context of robust machine learning systems under data drift and noise.
technical
02
Characterizing data drift
covariate shift, label shift, concept drift, and prior probability shift with formal definitions and real-world examples.
Characterizingcovariate
03
Characterizing noise
label noise, feature noise, adversarial noise, and annotation errors in supervised and unsupervised settings.
Characterizingadversarial
04
Drift detection fundamentals
statistical tests, windowing strategies, and sequential change-point detection for streaming data.
detectionstatistical
05
Noise detection and quantification
confusion matrix-based methods, loss distribution analysis, and robust statistics for outlier identification.
detectionquantification
06
Data quality assessment pipelines
profiling, validation, and monitoring for drift and noise in production ML systems.
assessmentpipelines
07
Robust loss functions
Huber, Tukey, quantile, and correntropy losses for training under label and feature noise.
functionsquantile
08
Regularization and constraints for robustness
weight decay, dropout, spectral normalization, and Lipschitz continuity.
Regularizationconstraints
09
Ensemble methods for drift and noise
bagging, boosting, random forests, and snapshot ensembles under distribution shift.
Ensembleboosting
10
Domain adaptation techniques
instance reweighting, feature alignment, and adversarial domain adaptation for covariate shift.
adaptationtechniques
11
Domain generalization
invariant risk minimization, causal representation learning, and meta-learning for unseen domains.
generalizationinvariant
12
Test-time adaptation
entropy minimization, self-training, and normalization layer adaptation under distribution shift.
Test-timeadaptation
13
Continual learning under concept drift
elastic weight consolidation, replay buffers, and dynamic architecture expansion.
Continualconsolidation
14
Online learning and adaptive algorithms
regret bounds, passive-aggressive updates, and drift-aware learning rates.
adaptivealgorithms
15
Label noise correction
noise transition matrix estimation, loss correction, and sample reweighting.
correctiontransition
16
Semi-supervised and self-supervised learning for noisy data
consistency regularization, contrastive learning, and pseudo-labeling.
Semi-supervisedself-supervised
17
Adversarial robustness
adversarial training, certified defenses, and randomized smoothing under input perturbations.
Adversarialrobustness
18
Uncertainty quantification for robust decision-making
Bayesian neural networks, deep ensembles, and conformal prediction.
Uncertaintyquantification
19
Out-of-distribution detection
density estimation, distance-based methods, and energy-based scores.
Out-of-distributiondetection
20
Anomaly detection in noisy and drifting data streams
isolation forests, autoencoders, and one-class SVMs.
detectiondrifting
21
Evaluation metrics for robustness
accuracy under shift, noise-robust metrics, and calibration measures.
Evaluationrobustness
22
Benchmark datasets and testbeds for drift and noise
CIFAR-10-C, DomainBed, WILDS, and streaming benchmarks.
Benchmarkdatasets
23
MLOps for drift and noise
monitoring, alerting, retraining triggers, and continuous evaluation pipelines.
MLOpsmonitoring
24
Case study
robust fraud detection under evolving transaction patterns and label noise.
detectionevolving
25
Case study
autonomous driving perception under weather, lighting, and sensor noise.
autonomousperception
26
Case study
medical diagnosis with noisy labels and domain shift across hospitals.
diagnosishospitals
27
Case study
natural language processing under dialectal variation, typos, and adversarial text perturbations.
languageprocessing
28
Scalable system design
distributed training, federated learning, and edge deployment for robust ML.
Scalabledistributed
29
Ethical and fairness considerations
bias amplification under drift, noise-induced disparities, and robust governance.
fairnessconsiderations
30
Capstone project
end-to-end design, implementation, and evaluation of a robust ML system under data drift and noise.
Capstoneend-to-end