Disentangling impact of capacity, objective, batchsize, estimators, and step-size on flow VI

Abhinav Agrawal, Justin Domke
Proceedings of The 28th International Conference on Artificial Intelligence and Statistics, PMLR 258:325-333, 2025.

Abstract

Normalizing flow-based variational inference (flow VI) is a promising approximate inference approach, but its performance remains inconsistent across studies. Numerous algorithmic choices influence flow VI’s performance. We conduct a step-by-step analysis to disentangle the impact of some of the key factors: capacity, objectives, gradient estimators, number of gradient estimates (batchsize), and step-sizes. Each step examines one factor while neutralizing others using insights from the previous steps and/or using extensive parallel computation. To facilitate high-fidelity evaluation, we curate a benchmark of synthetic targets that represent common posterior pathologies and allow for exact sampling. We provide specific recommendations for different factors and propose a flow VI recipe that matches or surpasses leading turnkey Hamiltonian Monte Carlo (HMC) methods.

Cite this Paper

BibTeX
@InProceedings{pmlr-v258-agrawal25a, title = {Disentangling impact of capacity, objective, batchsize, estimators, and step-size on flow VI}, author = {Agrawal, Abhinav and Domke, Justin}, booktitle = {Proceedings of The 28th International Conference on Artificial Intelligence and Statistics}, pages = {325--333}, year = {2025}, editor = {Li, Yingzhen and Mandt, Stephan and Agrawal, Shipra and Khan, Emtiyaz}, volume = {258}, series = {Proceedings of Machine Learning Research}, month = {03--05 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v258/main/assets/agrawal25a/agrawal25a.pdf}, url = {https://proceedings.mlr.press/v258/agrawal25a.html}, abstract = {Normalizing flow-based variational inference (flow VI) is a promising approximate inference approach, but its performance remains inconsistent across studies. Numerous algorithmic choices influence flow VI’s performance. We conduct a step-by-step analysis to disentangle the impact of some of the key factors: capacity, objectives, gradient estimators, number of gradient estimates (batchsize), and step-sizes. Each step examines one factor while neutralizing others using insights from the previous steps and/or using extensive parallel computation. To facilitate high-fidelity evaluation, we curate a benchmark of synthetic targets that represent common posterior pathologies and allow for exact sampling. We provide specific recommendations for different factors and propose a flow VI recipe that matches or surpasses leading turnkey Hamiltonian Monte Carlo (HMC) methods.} }
Endnote
%0 Conference Paper %T Disentangling impact of capacity, objective, batchsize, estimators, and step-size on flow VI %A Abhinav Agrawal %A Justin Domke %B Proceedings of The 28th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2025 %E Yingzhen Li %E Stephan Mandt %E Shipra Agrawal %E Emtiyaz Khan %F pmlr-v258-agrawal25a %I PMLR %P 325--333 %U https://proceedings.mlr.press/v258/agrawal25a.html %V 258 %X Normalizing flow-based variational inference (flow VI) is a promising approximate inference approach, but its performance remains inconsistent across studies. Numerous algorithmic choices influence flow VI’s performance. We conduct a step-by-step analysis to disentangle the impact of some of the key factors: capacity, objectives, gradient estimators, number of gradient estimates (batchsize), and step-sizes. Each step examines one factor while neutralizing others using insights from the previous steps and/or using extensive parallel computation. To facilitate high-fidelity evaluation, we curate a benchmark of synthetic targets that represent common posterior pathologies and allow for exact sampling. We provide specific recommendations for different factors and propose a flow VI recipe that matches or surpasses leading turnkey Hamiltonian Monte Carlo (HMC) methods.
APA
Agrawal, A. & Domke, J.. (2025). Disentangling impact of capacity, objective, batchsize, estimators, and step-size on flow VI. Proceedings of The 28th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 258:325-333 Available from https://proceedings.mlr.press/v258/agrawal25a.html.

Related Material