Statements (45)
Predicate | Object |
---|---|
gptkbp:instanceOf |
Machine Learning Paradigm
|
gptkbp:advantage |
Enables pretraining of models
Improves generalization Reduces need for labeled data |
gptkbp:distinctFrom |
Supervised Learning uses labeled data
Unsupervised Learning does not use labels or pseudo-labels |
gptkbp:example |
gptkb:Masked_Language_Modeling
gptkb:BYOL Autoencoders MoCo SimCLR Contrastive Learning |
gptkbp:field |
gptkb:Machine_Learning
gptkb:artificial_intelligence |
gptkbp:goal |
Learn representations from unlabeled data
|
https://www.w3.org/2000/01/rdf-schema#label |
Self-supervised Learning
|
gptkbp:method |
Generate pseudo-labels from data itself
|
gptkbp:notableFor |
gptkb:BERT
gptkb:Vision_Transformers GPT (Generative Pre-trained Transformer) Self-supervised Speech Models |
gptkbp:originatedIn |
Early 2010s
|
gptkbp:output |
Learned Representations
|
gptkbp:popularizedBy |
gptkb:Yann_LeCun
gptkb:Facebook_AI_Research |
gptkbp:relatedConcept |
gptkb:Representation_Learning
Data Augmentation Contrastive Loss Masked Prediction Pretext Task Pseudo-labeling Self-labeling |
gptkbp:relatedTo |
gptkb:Unsupervised_Learning
Supervised Learning |
gptkbp:trainer |
Unlabeled Data
|
gptkbp:usedFor |
Transfer Learning
Feature Extraction Pretraining Neural Networks Downstream Tasks |
gptkbp:usedIn |
gptkb:Computer_Vision
gptkb:Representation_Learning gptkb:Natural_Language_Processing gptkb:Speech_Recognition |
gptkbp:bfsParent |
gptkb:Foundation_Model
|
gptkbp:bfsLayer |
6
|