AI 용어집
인공지능 완전 사전
Masked Prediction
Technique consisting of randomly masking portions of the input data and training the model to predict the missing information.
Proxy Image Rotation
Pretext task where the model must predict the rotation angle applied to an image, forcing the learning of semantic spatial features.
Self-supervised Jigsaw Puzzle
Proxy task requiring the reassembly of shuffled image pieces, encouraging the learning of spatial structure and contextual relationships.
Temporal Context Prediction
Self-supervised method predicting the temporal order or context of sequential events without explicit annotations.
Self-supervised Inpainting
Reconstruction technique where the model fills masked regions of an image by learning contextual structures and textures.
Constrained Clustering
Self-supervised approach using automatically generated constraints to guide the clustering of unlabeled data.
Self-supervised Distillation
Process where a self-supervised teacher model transfers its knowledge to a more compact student model without manual labels.
Proxy Multimodal Alignment
Self-supervised task learning to align different modalities (text, image, sound) by exploiting their natural co-occurrences.
Positive Pair Generation
Process automatically creating similar examples for contrastive learning from a single unlabeled data source.
Neighborhood Prediction
Proxy task where the model predicts the nearest neighbors of a sample in feature space without labels.
Adversarial Self-Supervised Augmentation
Technique generating challenging but semantically preserved data augmentations to strengthen representation robustness.
Dynamic Contrastive Memory
Mechanism dynamically storing negative representations for contrastive learning without requiring large data batches.
Iterative Pseudo-Labeling
Self-supervised process progressively generating confidence labels to guide training on initially unlabeled data.