AI-ordlista
Den kompletta ordlistan över AI
Denoising Diffusion Probabilistic Models (DDPM)
Core diffusion architecture based on progressive noise addition and its inversion learning via neural networks
Latent Space Diffusion
Approach combining autoencoders and diffusion to operate in reduced-dimensional spaces, reducing computational costs
Score Matching and Score-Based Models
Underlying mathematical theory where models learn the gradient of the data distribution rather than the distribution itself
Stochastic vs Deterministic Diffusion
Distinction between probabilistic (stochastic) approaches and ODE solvers (deterministic) for the reverse diffusion process
Sampling Acceleration
Techniques to reduce the number of denoising steps required, including distillation methods and advanced schedulers
Conditional Diffusion
Extension allowing generation to be guided by specific conditions such as text, images, or other modalities
3D and Geometric Diffusion
Application of diffusion to the generation of 3D shapes, meshes, and point clouds with preservation of geometric constraints
Video and Temporal Diffusion
Spatio-temporal extension of diffusion for generating coherent video sequences with temporal dependency modeling
Audio and Wave Diffusion
Adaptation of diffusion for the generation and manipulation of audio waves, including speech and music
Hybrid Diffusion-Transformer Models
Architecture combining transformer attention with the generative power of diffusion for structured data
Classifier Guidance and Classifier-Free Guidance
Generation control techniques using either external classifiers or internal gradients to steer diffusion
Fokker-Planck Equation Theory
Mathematical foundation describing the evolution of probability densities in continuous diffusion processes
Diffusion for Inpainting and Editing
Specialized application for completing or modifying image regions while preserving overall coherence
Multi-Modal Diffusion
Coordinated generation across different modalities (text-image-audio) using shared diffusion processes
Optimization and Computational Efficiency
Techniques for reducing the memory and computational footprint of diffusion models for practical deployment