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

Sana Tonekaboni, Tina Behrouzi, Addison Weatherhead, Emily Fox, David Blei, Anna Goldenberg
Proceedings of the Forty-first Conference on Uncertainty in Artificial Intelligence, PMLR 286:4227-4250, 2025.

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.

Cite this Paper

BibTeX
@InProceedings{pmlr-v286-tonekaboni25a, title = {HDP-Flow: Generalizable Bayesian Nonparametric Model for Time Series State Discovery}, author = {Tonekaboni, Sana and Behrouzi, Tina and Weatherhead, Addison and Fox, Emily and Blei, David and Goldenberg, Anna}, booktitle = {Proceedings of the Forty-first Conference on Uncertainty in Artificial Intelligence}, pages = {4227--4250}, year = {2025}, editor = {Chiappa, Silvia and Magliacane, Sara}, volume = {286}, series = {Proceedings of Machine Learning Research}, month = {21--25 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v286/main/assets/tonekaboni25a/tonekaboni25a.pdf}, url = {https://proceedings.mlr.press/v286/tonekaboni25a.html}, 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.} }
Endnote
%0 Conference Paper %T HDP-Flow: Generalizable Bayesian Nonparametric Model for Time Series State Discovery %A Sana Tonekaboni %A Tina Behrouzi %A Addison Weatherhead %A Emily Fox %A David Blei %A Anna Goldenberg %B Proceedings of the Forty-first Conference on Uncertainty in Artificial Intelligence %C Proceedings of Machine Learning Research %D 2025 %E Silvia Chiappa %E Sara Magliacane %F pmlr-v286-tonekaboni25a %I PMLR %P 4227--4250 %U https://proceedings.mlr.press/v286/tonekaboni25a.html %V 286 %X 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.
APA
Tonekaboni, S., Behrouzi, T., Weatherhead, A., Fox, E., Blei, D. & Goldenberg, A.. (2025). HDP-Flow: Generalizable Bayesian Nonparametric Model for Time Series State Discovery. Proceedings of the Forty-first Conference on Uncertainty in Artificial Intelligence, in Proceedings of Machine Learning Research 286:4227-4250 Available from https://proceedings.mlr.press/v286/tonekaboni25a.html.

Related Material