This is the implementation of my medium article "Deep Auto-Encoders for Clustering: Understanding and Implementing in PyTorch".
This repository showcases how to apply Auto-Encoders to learn a lower-dimensional and easier separable presentation of the data. This is especially useful for tasks like clustering. To this end, the well-known MNIST dataset that comprises images of handwritten digits is used. The Auto-Encoder is implemented using PyTorch and scikit-learn is used to apply clustering.
The notebook shows that the accuracy of k-Means clustering can be improved by more than 20%-points using Auto-Encoders!