LAAT: Locally Aligned Ant Technique for discovering multiple faint low dimensional structures of varying density

Dimensionality reduction and clustering are often used as preliminary steps for many complex machine learning tasks. The presence of noise outliers can deteriorate the performance such preprocessing therefore impair subsequent analysis tremendously. In manifold learning, several studies indicate solutions removing background or close to structure when density is substantially higher than that exhibited by noise. However, in applications, including astronomical datasets, varies alongside manifolds buried a noisy background. We propose novel method extract based on idea Ant colony optimization. contrast existing random walk solutions, our technique captures points locally aligned with major directions manifold. Moreover, we empirically show biologically inspired formulation ant pheromone reinforces this behavior enabling it recover multiple embedded extremely data clouds. algorithm comparison state-of-the-art approaches detection demonstrated, synthetic real an N-body simulation cosmological volume.