Deep Embedded Clustering (DEC)

This is simplified pytorch-lightning implementation of 'Unsupervised Deep Embedding for Clustering Analysis' (ICML 2016).

Paper Review (Korean)

• [Post] Unsupervised Deep Embedding for Clustering Analysis

Environments

- torch == 1.5.0
- torchvision == 0.6.0
- pytorch-lightning == 0.8.5
- scipy == 1.5.2
- sklearn == 0.23.2

Result

• Standard K-Means Unsupervised Accuracy: 46%
• Initial Unsupervised Accuracy: 57.2%
• Trained Unsupervised Accuracy: 73.0%

p.s) Not follow paper settings strictly.

Dataset

• MNIST in torchvision

References

• [Paper] Unsupervised Deep Embedding for Clustering Analysis
• [Code] dec
• [Code] pt-dec