HDP-Flow: Generalizable Bayesian Nonparametric Model for Time Series State Discovery

Sana Tonekaboni, Tina Behrouzi, Addison Weatherhead, Emily Fox, David Blei, Anna Goldenberg

Keywords: Bayesian Nonparametric, Non-Stationary Time Series, Latent State Modeling, Dynamic State Discovery, Probabilistic Modeling

TL;DR: HDP-Flow is a Bayesian nonparametric model for unsupervised state discovery in dynamic, non-stationary time series, integrating BNP flexibility and deep generative models to learn evolving states in complex real-world data.

Abstract: We introduce HDP-Flow, a Bayesian nonparametric (BNP) model for unsupervised state discovery in dynamic, non-stationary time series data. Unlike prior work that assumes fixed states, HDPFlow models evolving datasets with unknown and variable latent states. By integrating the adaptability of BNP models with the expressive power of normalizing flows, HDP-Flow effectively models dynamic, non-stationary patterns, while learning transferable states across datasets with wellcalibrated uncertainty. We propose a scalable variational algorithm to enable efficient inference, addressing the limitations of traditional sampling-based BNP methods. HDP-Flow outperforms existing approaches in latent state identification and provides probabilistic insight into state distributions and transition dynamics. Evaluating HDP-Flow across two wearable datasets demonstrates transferability of states across diverse sub-populations, validating its robustness and generalizability.

Readers: auai.org/UAI/2025/Conference, auai.org/UAI/2025/Conference/Area_Chairs, auai.org/UAI/2025/Conference/Reviewers, auai.org/UAI/2025/Conference/Submission635/Authors, auai.org/UAI/2025/Conference/Submission635/Reproducibility_Reviewers

Submission Number: 635