AstroAI Workshop 2025
Simone Vilardi
Accelerating LAAT with Hub Pre-Processing: Efficient Detection of Filaments and Streams
Presenter: Simone Vilardi
Title: Accelerating LAAT with Hub Pre-Processing: Efficient Detection of Filaments and Streams
Date/Time: Monday, July 7th, 3:30 - 5:00 PM
Abstract: Finding cosmological structures such as filaments, streams, and clusters within the cosmic web is paramount to understanding the Large-Scale structure of the Universe. However, detecting and analysing such structures is challenging because they are often embedded in noisy, multi-dimensional data sets from large astronomical surveys or complex N-body simulations. The recently proposed Locally Aligned Ant Technique (LAAT) provides an efficient, biologically inspired, agent-based tool in which autonomous ‘ants’ make local decisions to highlight and recover multiple faint and noisy structures of potentially different dimensionality and density despite large amounts of background noise. Some cosmological features, such as nodes in the cosmic web or globular clusters, act as hubs, containing thousands of times more particles in a small region than the surrounding filamentary and stream-like structures. These high-density regions are located extremely quickly by the ants and are repeatedly highlighted throughout the runtime of the algorithm, leading to unnecessary computational overhead. Therefore, we present a two-stage approach: in the first stage, a fast preprocessing step identifies such hubs and replaces them with a tailored likelihood model; in the second stage, a mixed likelihood-pheromone scheme guides the ant movement, allowing more efficient bridging between dense regions and accelerating the detection of filaments and streams. This modification addresses the bottlenecks of the original LAAT and improves both its efficiency and robustness in recovering the large-scale cosmological web.