12:00–1:00 pm ERC 401 & zoom
Host: Vikram Dwarkadas
-
Dimitrios Tanoglidis "Bayesian Neural Networks, or Neural Networks that can say 'I don’t know'!"
Deep Learning is revolutionizing our life and the way we do science, including astrophysics. However, the majority of neural network models currently used are deterministic: their output prediction is a single number without any estimate of the uncertainty in that prediction. In this talk, I will discuss the different types of errors present in deep learning models and I will give a short overview of the methods we have to quantify them. Then, I will focus on Bayesian Neural Networks (BNNs), a class of probabilistic network models that are suitable for capturing uncertainty and thus promising for scientific applications. Finally, I will briefly present my ongoing work on using BNNs to measure structural parameters of low-surface-brightness galaxies (LSBGs).
-
Patricio Gallardo "A story of a pandemic in two acts: optical design for CMB-S4 and measuring galaxy clusters motions"
In this talk I will explore two somewhat disjoint topics related to CMB instrumentation and CMB science: 1.- I will talk about the optical design of the large aperture telescope cameras for CMB-S4, the next generation CMB experiment. These arrays of cameras will populate the CMB-S4 large-aperture telescopes' focal planes allowing observations with hundreds of thousands of detectors to achieve unprecedented levels of sensitivity. 2.- I will also discuss measurements of the motions of galaxy clusters using the pairwise kinematic SZ effect.
