Explainable Novel Category Discovery in Semantic Concept Space

2026-07-05

AI Summary

This work addresses the lack of semantic interpretability in existing novel class discovery methods by embedding the task into a structured semantic concept space. Leveraging pretrained multimodal models, it aligns visual–language similarity priors to learn concept representations from unlabeled data and jointly optimizes self-labeling objectives for both labeled and unlabeled samples in the concept-space logits. The approach inherently endows each discovered class with human-interpretable explanations composed of stable concept signatures and instance-level evidence. Theoretically, it constrains the hypothesis space and favors semantically coherent partitions. Experiments demonstrate state-of-the-art performance on CIFAR-10 (92.63%), CIFAR-100 (76.45%), and CUB-200, with the added distinction of being the only method that provides both cluster-level and instance-level human-readable interpretations in novel class discovery.

Abstract

Novel category discovery aims to identify unseen classes from unlabeled data by transferring knowledge from labeled categories, but most existing methods perform discovery in opaque latent feature spaces. As a result, they may separate novel categories accurately while providing little insight into what semantic evidence defines each discovered group. We propose xNCD, an explainable novel category discovery framework that performs both representation-based discovery and pseudo-label assignment directly in a structured semantic concept space. Instead of clustering arbitrary deep features, xNCD learns a label-free concept representation by aligning visual features with vision-language similarity priors from pretrained multimodal models, and then applies a unified labeled-and-unlabeled self-labeling objective over concept-space logits. This design makes each discovered category explainable by construction through stable concept signatures and instance-level concept evidence. Theoretically, we show that routing discovery through a semantic concept bottleneck induces a strict restriction of the feature-space hypothesis class, excluding a large family of unconstrained decision rules and biasing induced partitions toward semantically interpretable concept coordinates. Experiments on CIFAR-10, CIFAR-100, and CUB-200 demonstrate that xNCD preserves strong discovery performance while providing intrinsic explanations. Under task-agnostic evaluation, xNCD achieves 92.63% overall accuracy on CIFAR-10, close to UNO's 93.4%, and improves CIFAR-100 overall accuracy from 73.2% to 76.45%, while being the only compared method that provides human-readable cluster- and instance-level explanations.

Problem

• Explainable AI
• Novel Category Discovery
• Semantic Concept Space
• Unsupervised Learning
• Interpretability

Innovation

• Explainable AI
• Novel Category Discovery
• Semantic Concept Space
• Vision-Language Models
• Self-labeling