Weak Lensing (WL) surveys are reaching unprecedented depths, enabling the investigation of very small angular scales. At these scales, nonlinear gravitational effects lead to higher-order correlations making the matter distribution highly non-Gaussian. Extracting this information using traditional statistics has proven difficult, and Machine Learning based summary statistics have emerged as a powerful alternative. We explore the capabilities of a discriminative, Convolutional Neural Networks (CNN) based approach, focusing on parameter constraints in the ($Ω_m$, $σ_8$) cosmological parameter space. Leveraging novel training loss functions and network representations on WL mock datasets without baryons, we show that our models achieve $\sim 5$ times stronger constraints than the power spectrum, $\sim 3$ stronger constraints than peak counts, and $\sim 2$ stronger constraints than previous CNN-learned summary statistics and scattering transforms, for noise levels relevant to Rubin or Euclid. For WL convergence maps with baryonic physics, our models achieve $\sim 2.3$ times stronger constraining power than the power spectrum at these noise levels, also outperforming previous summary statistics. To further explore the possibilities of CNNs for this task, we also discuss transfer learning where we adapt pre-trained models, trained on different tasks or datasets, for cosmological inference, finding that these do not improve the performance.
Galaxy shape measurement with convolutional neural networks
我们从训练和评估卷积神经网络(CNN)来预测星系形状,从暗能量调查(DES DR 1)的第一次数据发布的宽视场调查图像中得到我们的结果。我们使用传统的形状测量作为地面实况,从重叠,更深的调查,较少的天空覆盖,加拿大-法国夏威夷望远镜透镜调查(CFHTLenS)。我们证明,CNN预测从单波段DES图像再现的结果CFHTLenS在明亮的幅度和显示更高的相关性与CFHTLenS在较暗的幅度比极大似然模型拟合估计的DES Y1 im 3shape目录。用CNN预测形状参数也非常快,每个星系只需要0.2毫秒,比前向模型拟合提高了4个数量级以上。CNN还可以在使用同一星系的多个图像时准确地预测形状,即使在不同的色带中,也没有额外的计算开销。CNN对微弱的天体也更精确,与DES Y1元校准目录相比,CNN的优势对于蓝色星系比红色星系更明显,DES Y1元校准目录使用riz波段图像拟合单一高斯轮廓。我们证明了在元校准自校准框架内的CNN形状预测产生具有可忽略的乘法偏差的剪切估计,$ m < 10^{-3}$,并且没有显著的PSF泄漏。我们提出的设置适用于当前和下一代弱透镜测量,其中可以在专用的深场中测量更高质量的地面实况形状。
Uncertainties in Parameters Estimated with Neural Networks: Application to Strong Gravitational Lensing
在Hezaveh et al. 2017中,我们证明了深度学习可用于模型参数估计和训练卷积神经网络,以确定强引力透镜系统的参数。在这里,我们展示了一种方法来获得这些参数的不确定性。我们审查变分推理的框架,以获得近似后验的贝叶斯神经网络,并将其应用到一个网络训练,估计参数的奇异等温椭球加上外部剪切和总通量放大。我们表明,该方法可以捕捉到由于输入数据中不同程度的噪声,以及由网络的训练和结构相关的错误的不确定性。为了评估由此产生的不确定性的准确性,我们计算每个透镜参数的边缘化分布的覆盖概率。通过调整一个超参数,辍学率,我们获得的覆盖概率近似等于他们计算的置信水平,从而产生准确和精确的不确定性估计。我们的研究结果表明,神经网络可以快速替代蒙特卡洛马尔可夫链参数不确定性估计在许多实际应用中,允许超过7个数量级的速度提高。
Deep Learning galaxy cluster's structural parameters from Weak Lensing observations
University of Ferrara(费拉拉大学);INAF – OAS, Osservatorio di Astrofisica e Scienza dello Spazio di Bologna(意大利国家天体物理研究所博洛尼亚天文与空间科学观测台);INFN-Sezione di Bologna(意大利国家核物理研究所博洛尼亚分部);Dipartimento di Fisica e Astronomia "Augusto Righi" - Alma Mater Studiorum Università di Bologna(博洛尼亚大学阿古斯托·里吉物理与天文学系);INAF-Osservatorio Astronomico di Capodimonte(意大利国家天体物理研究所卡波迪蒙特天文台);Argelander-Institut für Astronomie (AIfA), Universität Bonn(波恩大学阿尔加纳德天文学研究所)