IRIS imaging algorithm posterior samples for the DSHARP survey + Score Model weights.

Inferring sky surface brightness distributions from noisy interferometric data in a principled statistical framework has been a key challenge in radio astronomy. In this work, we introduce Imaging for Radio Interferometry with Score-based models (IRIS). We use score-based models trained on optical images of galaxies as an expressive prior in combination with a Gaussian likelihood in the uv-space to infer images of protoplanetary disks from visibility data of the DSHARP survey conducted by ALMA. We demonstrate the advantages of this framework compared with traditional radio interferometry imaging algorithms, showing that it produces plausible posterior samples despite the use of a misspecified galaxy prior. Through coverage testing on simulations, we empirically evaluate the accuracy of this approach to generate calibrated posterior samples. Our code is open source and freely available at https://github.com/EnceladeCandy/IRIS. This dataset includes 250 posterior samples for each protoplanetary disk from the DSHARP survey and the neural network weights for the different score-based models used in this work (ADD LINK TO ARXIV PAPER). Search for the notebook "3_zenodo_data.ipynb" in the tutorials folder of the github repository to better understand the structure of the dataset. Each protoplanetary disk has an associated .h5 file containing the posterior samples obtained with the VE PROBES prior. The RU Lup disk has different .h5 files, obtained with different priors. The filename contains the prior used. The HD 143006 disk has different .h5 files obtained with the VP SKIRT prior but different scaling factor. The scaling factor is in the filename in each case (or can be retrieved inside the .h5) file.The Elias 24 disk has different .h5 files, obtained by changing the sampling procedure with the VE PROBES prior. The sampler ("euler" or "pc") is indicated in the filename. The specific parameters for the pc sampler can be found with the sampling_params key in the dataset. For the Euler sampler, the number of steps was fixed at 4000 steps.在严谨规范的统计框架下，从含噪干涉测量数据中推断射电天空表面亮度分布，始终是射电天文学领域的核心挑战。本研究提出了基于得分匹配模型（score-based models）的射电干涉成像方法（Imaging for Radio Interferometry with Score-based models，以下简称IRIS）。我们将在星系光学图像上训练得到的得分匹配模型作为具有强表达能力的先验分布，结合uv空间中的高斯似然函数，从ALMA开展的DSHARP巡天的可见度数据中，推断原行星盘的射电图像。我们验证了该框架相较于传统射电干涉成像算法的优势，结果表明，即便使用了与真实分布失配的星系先验，该方法仍能生成合理的后验样本。通过对模拟数据开展覆盖性测试，我们从实验层面评估了该方法生成校准后验样本的准确性。本研究的代码已开源，可在https://github.com/EnceladeCandy/IRIS免费获取。本数据集包含DSHARP巡天中各原行星盘的250份后验样本，以及本研究中使用的各类得分匹配模型的神经网络权重（详见arxiv论文链接）。请在该GitHub仓库的教程文件夹中查找名为"3_zenodo_data.ipynb"的Notebook，以更好地理解数据集的结构。每个原行星盘对应一个.h5文件，其中存储了使用VE PROBES先验得到的后验样本。RU Lup原行星盘包含多份.h5文件，分别对应不同的先验设置，文件名中标注了所使用的先验信息。HD 143006原行星盘的.h5文件采用VP SKIRT先验，但使用了不同的缩放因子，缩放因子可在文件名中查看，或从.h5文件内部读取。Elias 24原行星盘的.h5文件通过调整VE PROBES先验下的采样流程生成，文件名中标注了采样器类型（"euler"或"pc"）。pc采样器的具体参数可通过数据集内的sampling_params键获取。对于Euler采样器，其采样步数固定为4000步。