KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
Forward Diffusion Process
Iterative process that progressively adds Gaussian noise to the original data until obtaining a purely random distribution, serving as the foundation for diffusion models in data augmentation.
Reverse Diffusion Process
Reverse process that learns to progressively denoise data to reconstruct or generate new variations, essential for creating realistic augmented samples.
Data Augmentation by Diffusion
Technique using diffusion models to generate structured and realistic variations of training data, improving the robustness and generalization of machine learning models.
Class-Conditional Diffusion
Extension of diffusion models incorporating class information to control the generation of augmentations specific to each category in the dataset.
Diffusion-Based Synthetic Data
Artificial data generated by diffusion models, preserving the statistical and structural characteristics of the original data while introducing controlled variation.
Variation Generation
Process of creating multiple variations of an original sample using different starting points in the diffusion space, thereby enriching the diversity of the augmented dataset.
Diffusion Sampling
Sampling methods of the reverse diffusion process, determining the quality and diversity of generated augmented data through strategies like DDIM or DPM-Solver.
Noise Prediction Network
Neural network trained to predict the noise added at each diffusion step, constituting the core of diffusion models for controlled generation of augmented data.
Diffusion Trajectory
Path taken by data in the diffusion space from the noisy state to reconstruction, directly influencing the nature and quality of the produced augmentations.
Augmented Dataset Generation
Systematic process of creating extended datasets using diffusion models, combining original data and synthetic variations to improve learning performance.
Diffusion-Based Feature Enhancement
Application of diffusion models to enhance or correct specific data features while preserving their overall semantic integrity.
Controlled Diffusion
Techniques guiding the diffusion process with specific constraints or conditions to generate targeted augmentations respecting certain desired properties.
Diffusion Interpolation
Method creating intermediate samples between two or more data points in the diffusion space, enabling progressive and controlled augmentation.
Progressive Noise Addition
Augmentation strategy adding noise progressively to data through the diffusion process, creating subtle to significant variations for robust training.
Diffusion-based Outlier Detection
Use of diffusion models to identify and generate examples of boundary or rare configurations, enhancing model resilience to extreme cases.