AI-ordlista
Den kompletta ordlistan över AI
Stochastic Transformation
Mathematical process applying progressive random transformations to data, enabling controlled transition between probability distributions in latent space.
Score-Based Model
Neural network architecture learning to predict the gradient of the log-probability potential field, used to reverse stochastic diffusion processes.
Reverse Diffusion Process
Regressive phase learning to iteratively denoise data, reconstructing the original sample from noise using estimated score gradients.
Fokker-Planck Equation
Partial differential equation describing the temporal evolution of probability density in diffusion processes, directly linking SDE to distributional dynamics.
Ornstein-Uhlenbeck Process
Stationary stochastic process with mean reversion, fundamentally used in diffusion models to control the dynamics of noise addition and removal.
Diffusion Rate
Scalar parameter controlling the intensity of noise added at each time step, determining the speed and stability of stochastic transformation.
Continuous Markov Chain
Continuous-time stochastic process where future states depend only on the present state, providing the mathematical foundation for differential diffusion models.
Score Function
Gradient of the logarithm of probability density with respect to data, pointing toward high-density regions and guiding noise deconstruction in generative models.
Continuous Diffusion Time
Normalized positive real parameter in [0,1] representing the continuous evolution of the stochastic process, enabling a unified differential formulation.
Estimated Score Gradients
Numerical approximations of log-likelihood gradients computed by neural networks, replacing inaccessible analytical gradients in complex distributions.