Few-shot Novel Category Discovery

2025-05-13

AI Summary

This paper addresses the limitation of inductive learning in open-set novel class discovery (NCD), where conventional transductive methods rely heavily on access to the entire unlabeled dataset. To overcome this, we propose few-shot novel class discovery (FSNCD)—a new paradigm enabling dynamic switching between supervised recognition of known classes and unsupervised clustering of unknown classes using only a few labeled support samples per novel class. This is the first work to integrate few-shot model adaptability into NCD, eliminating dependence on full unlabeled data. Methodologically, we introduce semi-supervised hierarchical clustering (SHC) and uncertainty-aware K-means (UKC), jointly leveraging prior-guided initialization, uncertainty calibration, and cross-task knowledge transfer. Extensive experiments across five benchmark datasets demonstrate consistent and significant improvements over state-of-the-art methods, particularly in generalization to and discrimination among unknown classes, under varying annotation budgets and task-switching configurations.

Technology Category

• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Data Mining & Knowledge Management: Anomaly/Outlier Detection
• Search and Optimization: Mixed Discrete/Continuous Search

Application Category

• Web Mining and Content Analysis: Normalization, clustering, classification, and summarization of Web text
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

The recently proposed Novel Category Discovery (NCD) adapt paradigm of transductive learning hinders its application in more real-world scenarios. In fact, few labeled data in part of new categories can well alleviate this burden, which coincides with the ease that people can label few of new category data. Therefore, this paper presents a new setting in which a trained agent is able to flexibly switch between the tasks of identifying examples of known (labelled) classes and clustering novel (completely unlabeled) classes as the number of query examples increases by leveraging knowledge learned from only a few (handful) support examples. Drawing inspiration from the discovery of novel categories using prior-based clustering algorithms, we introduce a novel framework that further relaxes its assumptions to the real-world open set level by unifying the concept of model adaptability in few-shot learning. We refer to this setting as Few-Shot Novel Category Discovery (FSNCD) and propose Semi-supervised Hierarchical Clustering (SHC) and Uncertainty-aware K-means Clustering (UKC) to examine the model's reasoning capabilities. Extensive experiments and detailed analysis on five commonly used datasets demonstrate that our methods can achieve leading performance levels across different task settings and scenarios.

Problem

• Addressing Novel Category Discovery with few labeled data
• Enabling flexible switching between known and novel class identification
• Improving clustering performance in real-world open-set scenarios

Innovation

• Few-shot learning for novel category discovery
• Semi-supervised hierarchical clustering approach
• Uncertainty-aware K-means clustering technique