Learning Method Papers

Learning method paper note는 objective, pretraining signal, transfer, fine-tuning, active learning, reinforcement learning, preference optimization, representation learning을 중심으로 봅니다.

Reading Axes

• What target signal is constructed?
• Is the method supervised, self-supervised, contrastive, masked, predictive, active, reinforcement-based, or preference-based?
• What changes: data construction, objective, architecture, optimization, or evaluation?
• Does the representation transfer to downstream tasks?
• Are train/test near-duplicates or benchmark leakage controlled?
• Is the claim about representation quality, transfer, data efficiency, stability, or preference alignment?

Concepts

• Modalities
• Multimodal learning
• Learning methods
• Self-supervised learning
• Masked modeling
• Contrastive learning
• JEPA
• Fine-tuning
• Active learning
• Reinforcement learning
• Reward modeling
• Preference optimization

Related

• Learning methods
• Leakage
• Claim extraction
• Paper comparison matrix
• Paper review workflow