Robust volatility updates for Hierarchical Gaussian Filtering

2026-05-05 · Source: cs.LG updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Mathematics & Computational Sciences, Health & Medical Research · Depth: Expert, extended

Summary

A new report introduces robust volatility updates for Hierarchical Gaussian Filtering (HGF) networks, addressing a critical issue where the original update equations for variance-targeting parents (volatility coupling) could lead to negative posterior precision, causing algorithmic termination. The modified approach, termed "unbounded HGF" (uHGF), employs an interpolated quadratic approximation to the variational energy, combining expansions at the prior prediction and a second mode located using the Lambert W function. This ensures positive posterior precision across the entire parameter space, even with large prediction errors. Comparative simulations demonstrate that uHGF successfully filters time series under conditions where the classic HGF crashes, improving approximation quality by almost two orders of magnitude (mean KL divergence of 0.023 vs. 1.34) and extending parameter space coverage from 77.3% to 100%. The solution also generalizes to the more complex Generalized HGF (gHGF) networks.

Key takeaway

For AI Scientists and Machine Learning Engineers developing or applying Hierarchical Gaussian Filtering models, adopting the new unbounded HGF (uHGF) update equations is crucial. This modification eliminates the risk of algorithmic crashes due to negative posterior precision, significantly expanding the usable parameter space and enabling more reliable parameter estimation. You should integrate the uHGF implementation into your HGF workflows to enhance model robustness and accuracy, particularly when dealing with high meta-volatility or large prediction errors.

Key insights

A new HGF update method ensures robust volatility estimation by preventing negative posterior precision across all parameter ranges.

Principles

• Variational energy must be quadratic for Gaussian posterior.
• Negative posterior precision is a logical impossibility.
• Interpolation improves approximation quality.

Method

The method interpolates between two quadratic expansions of variational energy: one at the prior prediction and another at a second mode found via the Lambert W function, then blends them using softmax-weighted moment matching.

In practice

• Use uHGF for robust HGF model fitting.
• Apply Lambert W function for mode location.
• Implement softmax blending for multi-modal approximations.

Topics

• Hierarchical Gaussian Filtering
• Variational Inference
• Volatility Estimation
• Quadratic Approximation
• Lambert W Function

Code references

• ComputationalPsychiatry/HierarchicalGaussianFiltering.jl
• ComputationalPsychiatry/pyhgf
• ComputationalPsychiatry/hgf-toolbox
• ilabcode/UnboundedHGF.jl

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• On the QUEST for Uncertainty Quantification via Highest Density Regions — QUEST offers a novel UQ framework using highest density regions to overcome pointwise predictive risk limitations.
• A Streaming Sparse Cholesky Method for Derivative-Informed Gaussian Process Surrogates Within Digital Twin Applications — Derivative-informed sparse Gaussian Processes enable accurate, dynamically updated digital twin surrogates for real-time system prognostics.
• The Score Kalman Filter — The Score Kalman Filter enables scalable nonlinear Bayesian filtering by replacing intractable partition function evaluations with efficient linear algebra.
• Optimal Stability of KL Divergence under Gaussian Perturbations — KL divergence stability under Gaussian perturbations extends to arbitrary distributions with optimal $\sqrt{\epsilon}$ degradation.
• Plug-in Losses for Evidential Deep Learning: A Simplified Framework for Uncertainty Estimation that Includes the Softmax Classifier — A simplified Evidential Deep Learning framework uses plug-in losses to achieve reliable, efficient uncertainty estimation, integrating softmax for broader applicability.
• HiAR: Efficient Autoregressive Long Video Generation via Hierarchical Denoising — Conditioning on same-noise-level context in autoregressive diffusion improves temporal consistency and mitigates error propagation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.LG updates on arXiv.org.