Deep Clustering of Compressed Variational Embeddings
This work addresses data compression and clustering challenges for applications with bandwidth and power constraints, but it appears incremental as it combines existing methods like VAEs and BMMs with a training trick.
The paper tackles the problem of clustering in compressed data domains to address limited communication bandwidth and low-power consumption, proposing a joint Variational Autoencoder with Bernoulli mixture model (VAB) that achieves compression and clustering simultaneously, enabling data vendors to benefit from security and bandwidth while consumers gain low computational complexity.
Motivated by the ever-increasing demands for limited communication bandwidth and low-power consumption, we propose a new methodology, named joint Variational Autoencoders with Bernoulli mixture models (VAB), for performing clustering in the compressed data domain. The idea is to reduce the data dimension by Variational Autoencoders (VAEs) and group data representations by Bernoulli mixture models (BMMs). Once jointly trained for compression and clustering, the model can be decomposed into two parts: a data vendor that encodes the raw data into compressed data, and a data consumer that classifies the received (compressed) data. In this way, the data vendor benefits from data security and communication bandwidth, while the data consumer benefits from low computational complexity. To enable training using the gradient descent algorithm, we propose to use the Gumbel-Softmax distribution to resolve the infeasibility of the back-propagation algorithm when assessing categorical samples.