Generating CMB Power Spectra via Artificial Neural Nets

The Cosmic Microwave Background (CMB) radiation is the oldest light in the universe and its temperature density fluctuations can reveal information about the early universe. CMB angular power spectra are typically generated with an Einstein-Boltzmann solver such as CLASS or CAMB. Replicating the spectra calculation can be done using very simple machine learning (ML) algorithms, notably multilayer perceptron regressors, and can offer a significant speed increase with little loss of accuracy. As a demonstration, CLASS was utilized to generate a small dataset upon which the ML model was created and tested. Comparing with the Planck CMB mission's sensitivities, the standard deviation of the differences between the ML model and the CLASS results are within 0.1 - 0.2 of Planck's experimental uncertainty on a per ℓ level for the C_ℓ^TT, C_ℓ^EE, and C_ℓ^TE power spectra while experiencing 𝒪(10⁴) speed gains.