ON THE PROVABLE ADVANTAGE OF UNSUPERVISED PRETRAINING

Keyphrases

• Unsupervised Pre-training 100%
• Downstream Task 57%
• Labeled Data 28%
• Generic Framework 28%
• Unlabeled Data 28%
• Class Function 14%
• Maximum Likelihood Estimation 14%
• Supervised Learning 14%
• Gaussian Mixture Model 14%
• Learning Task 14%
• Latent Variable Models 14%
• Machine Learning System 14%
• Empirical Risk Minimization 14%
• Excess Risk 14%
• Natural Approach 14%
• Factor Model 14%
• Theoretical Understanding 14%
• Contrastive Learning 14%
• Complexity Measures 14%
• Structural Assumption 14%
• Model Learning 14%
• Large-scale Machine Learning 14%
• Prediction Function 14%
• Abstract Classes 14%
• Unsupervised Representation Learning 14%

Computer Science

• Latent Variable Model 100%