Simulation-Based Inference and Unbinned Asimov Construction with Hybrid Neural Density EstimationNew

This paper addresses fundamental challenges in unbinned neural simulation-based inference (SBI) for high-dimensional parameter estimation. The authors propose hybrid neural density estimation that combines the strengths of neural ratio estimation—which excels at learning density ratios—with flow-based surrogate models that provide tractable density evaluation and efficient sampling. The method specifically tackles the limitation where learned ratios alone cannot produce normalized densities or generative models, while flow-based approaches suffer from residual estimation errors on complex implicit distributions. The Asimov construction technique is integrated to enable more robust unbinned inference with the hybrid approach.