|
Special Colloquium |
Alejandro Mus
| SCHEDULED |
Universitat de Valencia
In this talk, I present new frameworks for mm-VLBI data analysis,
which
include 1) A scattering mitigation framework for dynamic imaging of
time-variable or static objects and 2) A machine learning pipeline for
transient detection in VLBI data. We expect the combination of the two
frameworks would have a strong impact on a variety of science topics
like the long period repeating transient searches. First, I introduce
the advanced optimization techniques MOEA/D (Müller&Mus+2023,
Mus&Müller+2024) and Particle Swarm Optimization (Mus+2024) whose
flexibility allows the modeling for scattering mitigation based on
Stochastic Optics (Johnson16).Then, I present results on synthetic
movies at horizon scales, exploring different intrinsic structures,
source evolution and scatteringparameters (like the screen distance, and
speed). In a second step, I will show the performance of this framework
at 86 GHz, by using synthetic Sagittarius A* (SgrA*) Global mm-VLBI
Array(GMVA) data with various corruptions. We show that our algorithm
can recover the intrinsic SgrA* ring and the scattering screen, thanks
to the advantages of MOEA/D and PSO. Next, I will introduce a new
machine learning pipeline for detecting transients in VLBI data. Very
recently, pulsars with unexpected low period have been found (for
instance Caleb+2022,2024,Hurley-Walker+2023). We aim to revisit old data
using our fast and unsupervised algorithm to try to detect hidden slow
transient.
Finally, I will talk about ongoing work using full bandwidth
capabilities aim to mitigate scattering at lower frequencies, helping
the transient search with novel telescopes like next-generation VLA.