By name: hgf-robust-volatility-updates description: "Robust volatility updates for Hierarchical Gaussian Filtering (HGF). Improves stability and convergence of uncertainty estimation in perceptual inference. Activation: hierarchical gaussian filter, volatility update, perceptual inference, active inference, uncertainty estimation."
Robust Volatility Updates for Hierarchical Gaussian Filtering
Improved HGF volatility update rules for stable uncertainty estimation in hierarchical perceptual inference.
Metadata
- Source: arXiv:2605.04235
- Authors: Christoph Mathys, Nicolas Legrand, Peter Thestrup Waade, Nace Mikus, Lilian Aline Weber
- Published: 2026-05-07
- Categories: cs.LG, cs.NE, q-bio.NC, stat.ML
Core Methodology
HGF Volatility Robustness
- Standard HGF can exhibit instability in volatility estimation
- New update rules ensure bounded, well-behaved volatility estimates
- Maintains theoretical guarantees while improving numerical stability
- Compatible with existing HGF implementations (TAPAS, hgf R package)
Technical Framework
- Hierarchical Gaussian Filter: beliefs at multiple timescales
- Volatility level: estimates environmental change rate
- Robust updates: bounded influence functions prevent runaway estimates
- Convergence: proved stability under mild conditions
Applications
- Computational psychiatry (belief updating in patient populations)
- Decision-making under uncertainty
- Adaptive learning rate control
- fMRI/EEG model-based fMRI analysis
Related Skills
- free-energy-moe-routing
- online-generalised-predictive-coding
- neural-dynamics-decision-making