KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
One-Class Learning
Learning technique where the model is trained only on normal data to learn their distribution and identify deviations as anomalies.
Autoencoder Reconstruction Error
Measure quantifying the difference between original data and their reconstruction by an autoencoder, where a high error indicates a potential anomaly.
Support Vector Data Description
Variant of SVM that seeks a minimal hypersphere enclosing normal data, with exterior points considered anomalies.
Gaussian Mixture Model Anomaly
Probabilistic model representing data as a mixture of Gaussians, where low probabilities under the model indicate anomalies.
Pseudo-Labeling
Semi-supervised technique where the model generates labels for unlabeled data with high confidence, then using them for training.
Self-Training
Iterative approach where the model trains on its most reliable predictions to gradually expand the labeled training set.
Co-Training
Semi-supervised method using two classifiers on different views of data, training each other with their most confident predictions.
Graph-Based Anomaly Detection
Approach using graph structures to model relationships between points, with anomalies identified by their unusual connections.
Variational Autoencoder Anomaly
Use of VAE to model the distribution of normal data, with anomalies detected by their low probability under the learned model.
Consistency Regularization
Semi-supervised technique forcing the model to produce consistent predictions for different augmentations or perturbations of the same data.
Mean Shift Anomaly Detection
Non-parametric clustering algorithm identifying density modes, with points in low-density areas being classified as anomalies.