Andres Navarro

Andres Navarro

PhD researcher at University of Bonn, Exploiting ML techniques for Astrophysics research

Posts by Collection

projects

Compton Camera

Published:

To observe the inside of an object, we require high-energy photons such as X-rays and gamma radiation. Typically, imaging using X-rays or gamma radiation involves positioning the target between the radiation source and the detector. However, in many cases, it is not feasible to place the detector in the opposite direction to the source. Examples of such situations include detecting or imaging buried objects such as landmines, the inner side of pipes or walls, and others. The Compton Camera is a deviced developed by researchers in Bogotá and Darmstadt and lead by the Nuclear physics group of the Universidad Nacional de Colombia GFNUN, which uses backscattered gamma photons to produce images.

Propagation of PSF errors in cosmological estimates

Published:

Rho-statistics are the two-point correlation function (2PCF) of residual errors from the Point Spread Function (PSF) modeling. Stars are point sources, and the images we obtain from stars are actually a representation of the PSF. Therefore, we can use images of stars for modeling the PSF. The PSF has two main contributions: the optics of the telescope and the atmospheric smearing, and it is position and wavelength-dependent. The PSF needs to be interpolated to the galaxy’s position, and this interpolation might introduce errors given the variance of the PSF across the field of view. To study systematic biases that might arise from the interpolation scheme, a set of stars is left out from the PSF modeling. Then, by comparing the shape measurements of those reserved stars with the expected shape using the interpolation scheme, we can obtain residuals of the PSF that will help identify problematic sources, exposures, or regions where the PSF systematics exceed the requirements. Additionally, we can use these residuals as building blocks of PSF residual point correlation functions, which can be propagated in the likelihood analysis to study the consequences on final cosmological estimates.

MomentsML

Published:

MomentsML is a machine learning method designed to provide accurate shear estimates. It involves training a shallow neural network using realistic simulations with known shear values, to recover the shear signal. This is a challenging process given the multiple sources of noise and biases, such as PSF errors, blending of galaxies, detector effects, low signal-to-noise ratios, galaxy morphologies, and others.

publications

research

talks

teaching

Physics II

Undergraduate course, Universide Estadual de Campinas, Instituto de Física Gleb Wataghin, 2018


Nuclear Physics

Undergraduate course, Universide Estadual de Campinas, Instituto de Física Gleb Wataghin, 2019


Physics I

Undergraduate course, Universide Estadual de Campinas, Instituto de Física Gleb Wataghin, 2020