Generative Bayesian Filtering for State Estimation

2026-07-24 · Source: stat.ML updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, extended

Summary

Generative Bayesian Filtering (GBF) is a novel framework for inferring latent states from high-dimensional, noisy sensor signals in dynamic systems. It replaces traditional linear-Gaussian observation models, common in Kalman filters, with pretrained conditional generative models, specifically Conditional Variational Autoencoders (CVAE). GBF employs a Bayesian prediction-update recursion, where the measurement update is formulated as a score-based posterior sampling problem, combining dynamical priors with CVAE-induced likelihoods. This approach enhances flexibility and uncertainty quantification. Experiments on synthetic temporal MNIST and Fashion-MNIST datasets, along with real-world applications like Laser Powder Bed Fusion (LPBF) manufacturing monitoring and ECG-based arrhythmia diagnosis, demonstrate GBF's improved state estimation accuracy and robustness, particularly against noise and distribution shifts, outperforming several baseline methods.

Key takeaway

For Machine Learning Engineers developing robust state estimation systems from high-dimensional, noisy sensor data, consider Generative Bayesian Filtering (GBF). Its CVAE-based generative observation model and score-based posterior sampling offer superior accuracy and uncertainty quantification compared to traditional or discriminative methods, especially under distribution shifts. You should evaluate GBF for applications like industrial process monitoring or medical diagnostics to enhance reliability and adaptivity.

Key insights

GBF integrates CVAE-based generative models into Bayesian filtering for robust state estimation from high-dimensional, noisy sensor data.

Principles

• Generative models enhance filtering flexibility.
• Bayesian frameworks quantify state uncertainty.
• Temporal priors improve state adaptivity.

Method

GBF uses a CVAE decoder as a flexible likelihood model, then applies score-based posterior sampling via Stochastic Gradient Langevin Dynamics (SGLD) to combine dynamical priors with observations for recursive state updates.

In practice

• Monitor manufacturing processes for defects.
• Diagnose cardiac states from ECG signals.
• Track system states in noisy IoT environments.

Topics

• Bayesian Filtering
• Generative AI
• Conditional Variational Autoencoders
• State Estimation
• Score-based Sampling
• Process Monitoring
• Arrhythmia Diagnosis

Code references

• leicao25/Generative_Bayesian_Filtering

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

Related on AIssential

• A Perception vs. Distortion Perspective on Score-Based Generative Channel Estimation — Score-based channel estimation's efficacy is clarified by the perception-distortion tradeoff, especially under high uncertainty.
• GenEraser: Generalizable Video Object Removal via Balanced Text-Mask Guidance and Decoupled Locator-Preserver — GenEraser enhances video object removal by integrating multimodal guidance and a decoupled architecture to manage complex effects and preserve backgrounds.
• Generative Bayesian Inference with GANs — B-GAN bridges ABC and GANs to enable direct posterior sampling without explicit likelihoods.
• End-to-End Identifiable and Consistent Recurrent Switching Dynamical Systems — $\Omega$SDS offers exact likelihood optimization for identifiable recurrent switching dynamical systems, outperforming VAEs.
• Learning Probabilistic Filters with Strictly Proper Scoring Rules — The Proper Scoring Ensemble Filter (PSEF) learns robust Bayesian filters for complex dynamical systems using synthetic data and strictly proper scoring rules.
• The Score Kalman Filter — The Score Kalman Filter enables scalable nonlinear Bayesian filtering by replacing intractable partition function evaluations with efficient linear algebra.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

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