Pre-train, Prompt, Predict

Feel free to let us know the missing papers (issue or pull request)

We highly recommend using the interactive paperlist for prompt-based learning.

Typology of Prompting Methods

Pre-trained Models (Detailed Description)
- Left-to-right Language Model
- Masked Language Model
- Prefix Language Model
- Encoder-Decoder
Prompt Engineering (Detailed Description)
- Shape
  - Cloze prompt: [7] [53]
  - Prefix prompt: [17] [26]
- Human Effort
  - Hand-crated
  - Automated
    - Discrete: [24] [27]
    - Continuous: [29] [59]
Answer Engineering (Detailed Description)
- Shape
  - Token: [53] [7]
  - Span: [19] [22]
  - Sentence: [17] [29]
- Human Effort
  - Hand-crated: [2] [53]
  - Automated
    - Discrete: [24] [13]
    - Continuous: [28]
Multi-Prompt Learning (Detailed Description)
- Prompt Ensemble: [11] [57]
- Prompt Augmentation: [48] [47]
- Prompt Composition: [52]
- Prompt Decomposition: [53]
Prompt-based Training Strategies (Detailed Description)
- Parameter Updating
  - Promptless Fine-tuning
  - Tuning-free Prompting: [17] [7]
  - Fixed-LM Prompt Tuning: [29] [28]
  - Fixed-prompt LM Tuning: [26] [13]
  - Prompt+LM Tuning: [33] [37]
- Training Sample Size
  - Zero-shot: [34] [43]
  - Few-shot: [19] [27]
  - Full-data: [21] [29]

If you aim to use the picture from our github repo or survey, please state the source of these pictures.
These pictures are drawn by Weizhe and co-designed by Pengfei.