KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
Curse of Dimensionality
Phenomenon where data becomes sparse and difficult to analyze when the number of dimensions increases exponentially, making traditional anomaly detection methods ineffective.
Variational Autoencoders
Deep neural networks that learn a compressed representation of normal data, where anomalies present high reconstruction errors due to their low probability in the latent space.
Robust PCA
Variant of Principal Component Analysis resistant to outliers, using robust estimators to identify anomalies without them influencing the dimensional decomposition.
Subspace Clustering
Approach detecting anomalies in different dimensional subspaces, where a point can be normal in certain dimensions but abnormal in other combinations.
Deep SVDD
Extension of Support Vector Data Description using deep neural networks to learn a compact enclosing sphere of normal data in a non-linear feature space.
Projection Pursuit
Method seeking data projections that reveal interesting structures or anomalies, particularly useful for detecting non-obvious patterns in high dimension.
Random Projection
Dimensionality reduction technique that approximately preserves distances between points, used to accelerate anomaly detection while maintaining geometric properties.
DAGMM
Deep Autoencoding Gaussian Mixture Model combining autoencoders and Gaussian mixture models to capture complex distributions and detect anomalies in very high dimensional spaces.