1: \begin{abstract}

2: We present Neural Quantile Estimation (NQE), a novel Simulation-Based Inference (SBI) method based on conditional quantile regression.

3: NQE autoregressively learns individual one dimensional quantiles for each posterior dimension, conditioned on the data and previous posterior dimensions.

4: Posterior samples are obtained by interpolating the predicted quantiles using monotonic cubic Hermite spline, with specific treatment for the tail behavior and multi-modal distributions.

5: We introduce an alternative definition for the Bayesian credible region using the local Cumulative Density Function (CDF), offering substantially faster evaluation than the traditional Highest Posterior Density Region (HPDR).

6: In case of limited simulation budget and/or known model misspecification, a post-processing calibration step can be integrated into NQE to ensure the unbiasedness of the posterior estimation with negligible additional computational cost.

7: We demonstrate that NQE achieves state-of-the-art performance on a variety of benchmark problems.

8: %We demonstrate that with appropriate regularization, NQE tends to predict conservative results, mitigating the risk of making over-confident claims in the low simulation budget regime.

9: %Our method can be straightforwardly generalized to Sequential Neural Quantile Estimation (SNQE) when amortization is not required, for which we allocate new simulations within a truncated proposal defined by the broadened credible regions from previous iterations.

10: %, for which a conservative posterior estimation during the first iterations may help to improve convergence.

11: %We demonstrate that the proposed (S)NQE methods achieve state-of-the-art performance on a variety of benchmark problems.

12: \end{abstract}

13: