Constraining primordial non-Gaussianity using Neural Networks

We present a novel approach to estimate the value of primordial non-Gaussianity ($f_{\rm NL}$) parameter directly from the Cosmic Microwave Background (CMB) maps using a convolutional neural network (CNN). While traditional methods rely on complex statistical techniques, this study proposes a simpler approach that employs a neural network to estimate $f_{\rm NL}$. The neural network model is trained on simulated CMB maps with known $f_{\rm NL}$ in range of $[-50,50]$, and its performance is evaluated using various metrics. The results indicate that the proposed approach can accurately estimate $f_{\rm NL}$ values from CMB maps with a significant reduction in complexity compared to traditional methods. With $500$ validation data, the $f^{\rm output}_{\rm NL}$ against $f^{\rm input}_{\rm NL}$ graph can be fitted as $y=ax+b$, where $a=0.980^{+0.098}_{-0.102}$ and $b=0.277^{+0.098}_{-0.101}$, indicating the unbiasedness of the primordial non-Gaussianity estimation. The results indicate that the CNN technique can be widely applied to other cosmological parameter estimation directly from CMB images.