Glosarium AI
Kamus lengkap Kecerdasan Buatan
TimeGAN
GAN architecture specifically designed to generate realistic time series by capturing temporal dependencies and complex multidimensional distributions of sequential data.
Temporal Variational Autoencoder
Generative model that learns a continuous latent representation of time series and can generate new series by sampling from this latent space while preserving temporal structure.
Temporal Jittering
Augmentation technique that adds Gaussian noise to time points to simulate measurement variations and improve model robustness against temporal uncertainties.
Time Warping
Temporal warping method that slightly modifies the evolution speed of the time series to create variations while preserving fundamental patterns and overall characteristics.
Temporal SMOTE
Adaptation of the SMOTE algorithm for time series that generates synthetic samples by interpolating between neighboring time points while respecting temporal continuity and sequential constraints.
Temporal Diffusion
Generative approach that progressively applies noise to time series and then learns to reverse this process to generate new realistic series with consistent temporal characteristics.
Window Slicing
Augmentation method that extracts temporal sub-windows of different sizes and positions from original series to increase training data diversity while preserving local patterns.
Temporal Mixup
Regularization strategy that linearly combines two time series with a random mixing coefficient to create new synthetic samples while maintaining the temporal coherence of the sequences.
Synthetic Wavelets
Generation technique that uses wavelet decomposition to create new time series by recombining different wavelet coefficients with adjustable parameters to control frequency and temporal characteristics.
Recurrent GAN
GAN architecture integrating recurrent networks (LSTM/GRU) in the generator and discriminator to effectively capture long-term dependencies and sequential patterns in the generated time series.
Spline Interpolation
Augmentation method that uses spline functions to interpolate between existing time points, creating smooth and realistic series with continuous gradients preserving temporal dynamics.
Temporal Copulas
Statistical approach that models the dependency structure between different time points to generate synthetic series preserving temporal correlations and marginal distributions.
Temporal Scaling
Augmentation technique that multiplies the time series values by a random factor to simulate different amplitudes while preserving the fundamental shape and temporal patterns.
Temporal Cropping
Method that randomly selects continuous segments of time series to create new shorter sequences, thus increasing data diversity while preserving important local patterns.
Transformer-Based GAN
Generative architecture using Transformer attention mechanisms to capture long-distance dependencies and complex relationships in multivariate time series during synthetic generation.
Denoising Augmentation
Strategy that adds controlled noise to time series and then trains a model to reconstruct the original data, thus generating robust and slightly varied versions of the initial sequences.
Random Walk Synthesis
Generative method that uses random walk processes with parameters calibrated on existing data to create synthetic time series preserving basic statistical properties.
Temporal Conditional GAN
GAN variant that generates time series conditioned by specific metadata or initial states, allowing precise control over the characteristics of generated sequences.
Predictive Autoencoding
Model that learns to predict future time points while compressing past sequences, enabling generation of coherent continuations of existing time series.