DeepDPM: Deep Clustering With an Unknown Number of Clusters (2022-03-27T00:00:00.000000Z)

TL;DR

This work introduces an effective deep-clustering method that does not require knowing the value of $K$ as it infers it during the learning, and is the first to report the performance of such a method on ImageNet.

Abstract

Deep Learning (DL) has shown great promise in the unsupervised task of clustering. That said, while in classical (i.e., non-deep) clustering the benefits of the nonparametric approach are well known, most deep-clustering methods are parametric: namely, they require a predefined and fixed number of clusters, denoted by K. When $K$ is unknown, however, using model-selection criteria to choose its optimal value might become computationally expensive, especially in DL as the training process would have to be repeated numerous times. In this work, we bridge this gap by introducing an effective deep-clustering method that does not require knowing the value of $K$ as it infers it during the learning. Using a split/merge framework, a dynamic architecture that adapts to the changing $K$, and a novel loss, our proposed method outperforms existing nonparametric methods (both classical and deep ones). While the very few existing deep nonparametric methods lack scalability, we demonstrate ours by being the first to report the performance of such a method on ImageNet. We also demonstrate the importance of inferring $K$ by showing how methods that fix it deteriorate in performance when their assumed $K$ value gets further from the ground-truth one, especially on imbalanced datasets. Our code is available at https://github.com/BGU-CS-VIL/DeepDPM.

Authors

O. Freifeld

4 papers

M. Ronen

1 papers

Shahaf E. Finder

1 papers

TL;DR

Abstract

Authors

References76 items

Common Failure Modes of Subcluster-based Sampling in Dirichlet Process Gaussian Mixture Models - and a Deep-learning Solution

Sampling in Dirichlet Process Mixture Models for Clustering Streaming Data

DNB: A Joint Learning Framework for Deep Bayesian Nonparametric Clustering

Deep Clustering and Representation Learning that Preserves Geometric Structures

Nonparametric Object and Parts Modeling With Lie Group Dynamics

Learning To Classify Images Without Labels

You Only Train Once: Loss-Conditional Training of Deep Networks

Improved Baselines with Momentum Contrastive Learning

A Simple Framework for Contrastive Learning of Visual Representations

Graph Embedded Pose Clustering for Anomaly Detection

VSB-DVM: An End-to-End Bayesian Nonparametric Generalization of Deep Variational Mixture Model

Bayesian Adaptive Superpixel Segmentation

Deep Clustering by Gaussian Mixture Variational Autoencoders With Graph Embedding

Neural Clustering Processes

N2D: (Not Too) Deep Clustering via Clustering the Local Manifold of an Autoencoded Embedding

Video Face Clustering With Unknown Number of Clusters

Streaming Adaptive Nonparametric Variational Autoencoder

Unsupervised Pre-Training of Image Features on Non-Curated Data

Distributed MCMC Inference in Dirichlet Process Mixture Models Using Julia

Trajectories as Topics: Multi-Object Tracking by Topic Discovery

Deep Clustering for Unsupervised Learning of Visual Features

Deep Embedding for Determining the Number of Clusters

Deep Continuous Clustering

UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction

Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms

From Patches to Images: A Nonparametric Generative Model

Variational Deep Embedding: A Generative Approach to Clustering

Towards K-means-friendly Spaces: Simultaneous Deep Learning and Clustering

Efficient Global Point Cloud Alignment Using Bayesian Nonparametric Mixtures

Semantically-Aware Aerial Reconstruction from Multi-modal Data

Unsupervised Deep Embedding for Clustering Analysis

Small-variance nonparametric clustering on the hypersphere

Hierarchical Bayesian models for unsupervised scene understanding

Deep Learning with Nonparametric Clustering

Parallel Sampling of HDPs using Sub-Cluster Splits

Bayesian Nonparametric Intrinsic Image Decomposition

Nonparametric Clustering with Distance Dependent Hierarchies

The Dependent Dirichlet Process Mixture of Objects for Detection-free Tracking and Object Modeling

Parallel Sampling of DP Mixture Models using Sub-Cluster Splits

Memoized Online Variational Inference for Dirichlet Process Mixture Models

An Incremental DPMM-Based Method for Trajectory Clustering, Modeling, and Retrieval

From Deformations to Parts: Motion-based Segmentation of 3D Objects

The MNIST Database of Handwritten Digit Images for Machine Learning Research [Best of the Web]

Stochastic variational inference

Nonparametric discovery of activity patterns from video collections

Nonparametric learning for layered segmentation of natural images

An Analysis of Single-Layer Networks in Unsupervised Feature Learning

Dirichlet process mixture models on symmetric positive definite matrices for appearance clustering in video surveillance applications

Bayesian data analysis.

ImageNet: A large-scale hierarchical image database

Shared Segmentation of Natural Scenes Using Dependent Pitman-Yor Processes

Incremental learning of nonparametric Bayesian mixture models

Describing Visual Scenes Using Transformed Objects and Parts

Learning Multiscale Representations of Natural Scenes Using Dirichlet Processes

Two-View Motion Segmentation by Mixtures of Dirichlet Process with Model Selection and Outlier Removal

OPTIMOL: Automatic Online Picture Collection via Incremental Model Learning

Accelerated Variational Dirichlet Process Mixtures

Hierarchical Dirichlet Processes

Variational inference for Dirichlet process mixtures

RCV1: A New Benchmark Collection for Text Categorization Research

A Split-Merge Markov chain Monte Carlo Procedure for the Dirichlet Process Mixture Model

A Density-Based Algorithm for Discovering Clusters in Large Spatial Databases with Noise

A Database for Handwritten Text Recognition Research

Maximum likelihood from incomplete data via the EM - algorithm plus discussions on the paper

Mixtures of Dirichlet Processes with Applications to Bayesian Nonparametric Problems

A Bayesian Analysis of Some Nonparametric Problems

Monte Carlo Sampling Methods Using Markov Chains and Their Applications

Non-Parametric Clustering Using Deep Neural Networks

Scalable and Flexible Clustering of Grouped Data via Parallel and Distributed Sampling in Versatile Hierarchical Dirichlet Processes

The Manhattan Frame Model—Manhattan World Inference in the Space of Surface Normals

Streaming Variational Inference for Dirichlet Process Mixtures

Sampling in computer vision and Bayesian nonparametric mixtures

Visualizing Data using t-SNE

Combinatorial Stochastic Processes

Graphical models for visual object recognition and tracking

We set Ψ = I × 0.005 in all the datasets, where I denotes the identity matrix. Note that we used the same configuration for the three datasets