• Uses flow-based proposal distributions to adaptively focus sampling budgets on high-probability regions.
• Demonstrates lower variance estimators than classical PMC across synthetic and real Bayesian filtering tasks.