Verfasst von / Written by Sebastian F. Genter
K
Keras
High-level neural network API offering modular building blocks for deep learning. Key features include:
- Intuitive interface for fast prototyping
- Multi-backend support (TensorFlow as default)
- Built-in convolutional/recurrent layers
- Easy model export for production
Simplifies implementing CNNs, RNNs, and hybrid architectures through object-oriented design.
Kernel Support Vector Machines (KSVMs)
SVM variant using kernel functions to separate nonlinear data. Transforms inputs to higher dimensions via:
- Radial Basis Function (RBF)
- Polynomial kernels
- Sigmoid mappings
Maximizes margin between classes while tolerating misclassifications via soft margins.
keypoints
Distinctive image locations identified through feature detection algorithms. Applications include:
- Object tracking (SIFT features)
- Panorama stitching (ORB descriptors)
- Facial recognition (landmark points)
Characterized by invariant properties across scales/rotations.
k-fold cross validation
Robust evaluation technique splitting data into k equal partitions. Process:
- Train on k-1 folds
- Validate on remaining fold
- Repeat k times
- Average results
Reduces variance vs single split, especially valuable for small datasets.
k-means
Centroid-based clustering algorithm minimizing within-cluster variance. Steps:
- Initialize k centroids
- Assign points to nearest centroid
- Recalculate centroids
- Repeat until convergence
Requires predefined k, sensitive to initialization (use k-means++).
k-median
Clustering variant using Manhattan distance and median-based centroids. Advantages:
- Robust to outliers
- Handles noise better than k-means
- Suitable for grid-like data distributions
Computationally heavier due to sorting requirements.