Inference of the optical depth to reionization τ from Planck CMB maps with convolutional neural networks

Open Access

Table 2.

τ predictions from 10 000 Gaussian CMB + noise simulations generated with three different, fixed fiducial τ values.

Test on Gaussian simulations

	143 GHz			143 + 100 GHz			143 × 100 GHz
	Gaussian training			Gaussian training			HL likelihood
fiducial τ			σ(τ_NN)			σ(τ_NN)			σ(τ_HL)
0.05	0.0508	0.0059	0.0066	0.0503	0.0054	0.0057	0.0496	0.0046	0.0047
0.06	0.0608	0.0065	0.0067	0.0600	0.0056	0.0059	0.0596	0.0048	0.0048
0.07	0.0712	0.0067	0.0070	0.0702	0.0057	0.0063	0.0697	0.0048	0.0049

Notes. The results correspond to the Gaussian NN training on one and two channels, and the Bayesian inference with a power spectrum likelihood. We show the posterior mean τ_NN/HL and standard deviation σ_NN/HL(τ) averaged over all simulations, as well as the scatter of τ_NN/HL over all simulations.

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.