Supervised, semi-supervised, unsupervised, self-supervised

Supervised 지도학습

• All samples are labeled
• Requires a large amount of labeled data
• Cannot handle samples with unexpected patterns

---

Semi-supervised 준지도학습

• Only a small proportion of samples are labeled

---

Unsupervised 비지도학습

• No labeled samples

---

Self-supervised 자기지도학습

• No labeled samples (Unsupervised)

Steps of self-supervised training

1. Pretraining (Pretext task)

• with unlabeled data
• learn the feature representations of the data

2. Finetuning (Supervised downstream task)

• with labeled data