Spectra: Exact Component Transport for Test-Time Prior Adaptation in Simulation-Based Inference
Spectra adapts diffusion-based SBI models to changed priors at test time without rerunning simulations or retraining.
The paper introduces a closed-form score-transport method for structured prior changes using a frozen diffusion model. Its target is amortized simulation-based inference, where reusable posterior models are normally tied to the training prior. The authors report accurate adaptation under strong prior shifts across six SBI benchmarks, with low online sampling cost. Source: ArXiv · AI/CL/LG's note.
The paper introduces a closed-form score-transport method for structured prior changes using a frozen diffusion model. Its target is amortized simulation-based inference, where reusable posterior models are normally tied to the training prior. The authors report accurate adaptation under strong prior shifts across six SBI benchmarks, with low online sampling cost. Source: ArXiv · AI/CL/LG's note.
score 4