Thésards 3ème année: Romain Paviot & Qiufan Lin

Europe/Paris
Lien zoom: https://univ-amu-fr.zoom.us/j/96100395873?pwd=SnFGUmRObGFEYTJhblNja0pLV0lMZz09 ID de réunion : 961 0039 5873 Code: CPPM Ou par téléphone: Trouvez votre numéro local : https://univ-amu-fr.zoom.us/u/aRLypGmva ID de réunion : 961 0039 5873 Code secret : 896013
Description

This CPPM seminar is part of our series of talks from the CPPM 3rd year PhD students.
It will be happening remotely on zoom:

Lien zoom: https://univ-amu-fr.zoom.us/j/96100395873?pwd=SnFGUmRObGFEYTJhblNja0pLV0lMZz09
ID de réunion : 961 0039 5873
Code: CPPM

Ou par téléphone: 
Trouvez votre numéro local : https://univ-amu-fr.zoom.us/u/aRLypGmva
ID de réunion : 961 0039 5873
Code secret : 896013
    • 14:00 14:30
      Characterizing the growth of structures using the extended Baryon Oscillation Spectroscopic Survey (eBOSS). 30m

      I will overview the final cosmological constraints from the extended Baryon Oscillation Spectroscopic Survey (eBOSS). After five years of observing the spectra of nearly one million objects, eBOSS provided us with the largest density map of our Universe, from which we were able to measure the rates of expansion and growth of structures. I will focus particularly on the study of the luminous red galaxy (LRG) sample. Those measurement, combined with the analogous measurements of the quasar and emission line galaxy samples, are the best constraints up to date on dark energy and modified gravity models at redshift z = 0.7.

      Orateur: romain paviot (LAM)
    • 14:30 15:00
      Correcting galaxy photometric redshift estimation biases using Deep Learning neural networks 30m

      Deep Learning neural networks are powerful tools to capture information from input data, and have been increasingly applied in astrophysical studies. However, without proper treatments, data-driven algorithms such as neural networks are prone to overfitting on information that is correlated yet not causally related to certain tasks (e.g., systematics, the prior distribution of training data, etc.), and thus result in a biased output harmful for subsequent analyses. It is therefore essential to correct biases caused by such irrelevant information. Using galaxy photometric redshift estimation as an example, I will demonstrate the approaches that we exploit to tackle the two major forms of biases in the existing Deep Learning techniques, namely the redshift-dependent residual and the mode collapse. Experiments show that these approaches are effective and potentially useful in real astrophysical analyses. They are also meaningful in helping us understand the training of neural networks for general classification or regression problems in computer science applications.

      Orateur: Qiufan Lin (CPPM)