Probabilistic continual learning discovers new user intents reliably
Uncertainty-Aware Continual Learning for Open-World Intent Discovery Under an evolving Label Space
Machine LearningComputation and Language
Summary
Many intelligent systems need to understand new user goals as they change over time instead of only recognizing a fixed set. The authors developed a method that uses uncertainty to decide when new user intents have appeared, and it carefully adds new categories without forgetting old ones. Their approach clusters user requests and only accepts well-supported new groups, improving the system's ability to learn continuously. Experiments showed the method could adapt steadily to new intents while maintaining what it already knew.
What this means in practice
- •For voice assistant developers: Improve voice assistants to recognize and learn new user requests dynamically as they emerge without manual updates.$Commercial implications: Enables commercial voice assistants to expand recognized commands autonomously, enhancing user experience and reducing manual labeling.
- •For customer service chatbot teams: Add new inquiry types to chatbots on the fly by detecting and confirming novel customer intents for ongoing improvement.
Authors
Pisante Aida, Formentin Simone
Abstract
Real-world intelligent systems increasingly operate under open-world conditions, where user intents are not fixed or exhaustively known a priori and may evolve as new interaction patterns emerge. This paper proposes a unified uncertainty-aware probabilistic framework for continual new intent discovery under an evolving label space. Each utterance is encoded through an adaptive $β$-VAE into a latent mean, used for classification and density modelling and a posterior uncertainty estimate acting as a global reliability signal. Classifier confidence, posterior uncertainty and DP-GMM likelihood are combined through a multi-signal decision mechanism to distinguish known intents from potentially novel samples. Candidate novel instances are clustered through a density-based discovery module and only reliable clusters are promoted to new labels, enabling controlled label-space expansion. Replay and Elastic Weight Consolidation mitigate catastrophic forgetting and preserve previously acquired knowledge. The paper formalises continual intent discovery as a structured multi-phase open-world problem, introduces adaptive label-space expansion under stability--plasticity constraints and uses posterior uncertainty to regulate trusted-sample selection, pseudo-labelling, novelty admission and replay. Experiments show high novelty precision, stable adaptation across sequential phases and limited forgetting. Near-zero NMI and ARI indicate limited reconstruction of the complete fine-grained intent taxonomy, consistent with the framework's conservative promotion strategy. Qualitative analyses nevertheless reveal dense and locally coherent semantic clusters, showing that reliable novel structures can be discovered without exhaustive recovery of the underlying taxonomy.