Fast End-to-End Framework for Cosmological Parameter Inference from Cosmic Microwave Background Data Using Machine Learning

Precise estimation of cosmological parameters from the cosmic microwave background (CMB) remains a central goal of modern cosmology and a key test of inflationary physics. However, this task is fundamentally limited by strong foreground contamination, primarily from Galactic emissions. In this Letter, we introduce a fast simulation-based end-to-end pipeline combining the Analytical Blind Separation (ABS) method for foreground removal with a neural network (NN) framework for cosmological parameter inference. While both tools are individually established, their integration into a single computationally tractable pipeline has not previously been demonstrated. This is primarily enabled by the exceptional computational efficiency of ABS, which, for the first time, makes large-ensemble simulation-based training feasible in the presence of an explicit component separation step. As a proof of principle under controlled full-sky conditions, we assess this framework for the forthcoming LiteBIRD and PICO satellite missions, obtaining 1 σ errors of 0.0035 (LiteBIRD) and 0.0030 (PICO) for the optical depth τ , and 0.0056 (LiteBIRD) and 0.0015 (PICO) for the tensor-to-scalar ratio r . Recovered parameters are consistent with input values within 1 σ across most of the test parameter space, with LiteBIRD results consistent with mission forecasts. This demonstrates that the ABS–NN framework provides a computationally efficient and scalable solution for end-to-end CMB inference.