AI Briefing[NeurIPS 2021] Part 2: A Review of Deep Clustering Based on Variational Deep Embedding - LG AI Research BLOG
Key point
This article examines deep clustering techniques using Variational Autoencoder and the latest research based on VaDE.
Details
The Variational Autoencoder (VAE) method, which extracts data through unsupervised learning, is used in various fields such as anomaly detection, classification, and image segmentation. This review introduces two VAE-based deep clustering techniques presented at NeurIPS 2021.
First, we cover the basics of clustering. Spectral Clustering is a method that gathers similarly shaped data by measuring the distance between samples, but it has a computational cost problem of having to calculate an N×N distance matrix when there are N data points. On the other hand, K-means has high computational efficiency, but its performance can degrade when the data distribution is complex or there are missing values.
Based on Variational Deep Embedding (VaDE), a core study in deep clustering, we analyze the following two recent studies in depth.
- Multi-Facet Clustering Variational Autoencoder
- Deep Conditional Gaussian Mixture Model for Constrained Clustering
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.