Neural Autoregressive Flow Bayesian Model Averaging Research Artifacts

Research artifacts for the Neural Autoregressive Flow Bayesian Model Averaging (NAF-BMA) method, a variational inference framework for scalable Bayesian model averaging. The work, authored by Jiefu Zhou, was published in April 2026 and includes simulation and real-data studies demonstrating the method's performance against MCMC and VBMA benchmarks.

Use Cases

• Benchmarking the NAF-BMA method's posterior model probability estimates against MCMC and VBMA results from simulation studies.
• Analyzing the joint posterior distribution over model parameters and model evidences produced by the unified optimization scheme.
• Evaluating the framework's scalability and accuracy on real-data study results included in the artifacts.
• Reproducing experiments on posterior dependency and multimodality capture using the provided neural autoregressive flow implementations.

Strengths

• Method validated across extensive simulation and real-data studies, demonstrating consistent performance.
• Framework is general and modular, designed for a broad class of Bayesian models with minimal model-specific derivations.
• Artifacts are shared under a permissive CC-BY-4.0 license, facilitating reuse and reproduction.

Limitations

• The dataset is small at 1.9 MB, primarily containing code and documentation rather than large-scale data.
• No raw columnar data or sample rows are provided; the primary content is methodological research outputs.
• Real-data applications are described but the specific underlying datasets may not be included, limiting direct analysis.

Provenance

Source

Jiefu Zhou via figshare.

Collection Method

Generated as part of academic research proposing the NAF-BMA method, including simulation and application studies.

Freshness

Last updated in April 2026, indicating very recent research.