Clustering with Neural Network and Index

TL;DR

This work tackles the challenge of inductive clustering for non-convex data by introducing Clustering with Neural Network and Index (CNNI), which trains a neural network using an internal clustering evaluation index as the loss. The approach unifies a neural-network classifier with end-to-end index-based optimization, exemplified by the MMJ-SC index and SCI variants, and is enhanced by the SBSBP optimization strategy and a mini-batch variant. Key contributions include the CNNI architecture, the SBSBP algorithm, SCI V2, automatic model selection demonstrations, and MMJ-K-means loss formulations for both hard and soft clustering. The results suggest CNNI can yield inducible clustering with flexible model capacity and can handle non-convex geometries, marking a first parametric inductive clustering capability for such data and offering a path toward scalable, index-guided unsupervised learning.

Abstract

A new model called Clustering with Neural Network and Index (CNNI) is introduced. CNNI uses a Neural Network to cluster data points. Training of the Neural Network mimics supervised learning, with an internal clustering evaluation index acting as the loss function. An experiment is conducted to test the feasibility of the new model, and compared with results of other clustering models like K-means and Gaussian Mixture Model (GMM). The result shows CNNI can work properly for clustering data; CNNI equipped with MMJ-SC, achieves the first parametric (inductive) clustering model that can deal with non-convex shaped (non-flat geometry) data.

Figures (12)

• Figure 1: Three datasets for Experiment I
• Figure 2: Four Neural Networks
• Figure 3: Clustering results and decision boundaries of data 0
• Figure 4: Clustering results and decision boundaries of data 1
• Figure 5: Clustering results and decision boundaries of data 2

Theorems & Definitions (2)

• Definition 5.1
• Definition 5.2