Pavlou, Orestis

2022-10-14, Pavlou, Orestis, Efstathiou, Andreas, Papadopoulou Lesta, Vicky, Andreas A. Ioannides, Constantinos Kourouyiannis, Christodoulos Karittevlis, Lichan Liu, Ioannis Michos, Michalis Papadopoulos, Evangelos Papaefthymiou

In this article, we present a unified framework for the analysis and characterization of a complex system and demonstrate its application in two diverse fields: neuroscience and astrophysics. The framework brings together techniques from graph theory, applied mathematics, and dimensionality reduction through principal component analysis (PCA), separating linear PCA and its extensions. The implementation of the framework maps an abstract multidimensional set of data into reduced representations, which enable the extraction of its most important properties (features) characterizing its complexity. These reduced representations can be sign-posted by known examples to provide meaningful descriptions of the results that can spur explanations of phenomena and support or negate proposed mechanisms in each application. In this work, we focus on the clustering aspects, highlighting relatively fixed stable properties of the system under study. We include examples where clustering leads to semantic maps and representations of dynamic processes within the same display. Although the framework is composed of existing theories and methods, its usefulness is exactly that it brings together seemingly different approaches, into a common framework, revealing their differences/commonalities, advantages/disadvantages, and suitability for a given application. The framework provides a number of different computational paths and techniques to choose from, based on the dimension reduction method to apply, the clustering approaches to be used, as well as the representations (embeddings) of the data in the reduced space. Although here it is applied to just two scientific domains, neuroscience and astrophysics, it can potentially be applied in several other branches of sciences, since it is not based on any specific domain knowledge. Copyright

2024-04, Pavlou, Orestis, Efstathiou, Andreas, Papadopoulou Lesta, Vicky, Agapi Rissaki, D. Fotakis

he mid-infrared spectra of galaxies are rich in features such as the Polycyclic Aromatic Hydrocarbon (PAH) and silicate dust features which give valuable information about the physics of galaxies and their evolution. For example they can provide information about the relative contribution of star formation and accretion from a supermassive black hole to the power output of galaxies. However, the mid-infrared spectra are currently available for a very small fraction of galaxies that have been detected in deep multi-wavelength surveys of the sky. In this paper we explore whether Deep Generative Network methods can be used to reconstruct mid-infrared spectra in the 5–35μm range using the limited multi-wavelength photometry in ∼20 bands from the ultraviolet to the submillimeter which is typically available in extragalactic surveys. For this purpose we use simulated spectra computed with a combination of radiative transfer models for starbursts, active galactic nucleus (AGN) tori and host galaxies. We find that our method using Deep Generative Networks, namely Generative Adversarial Networks and Generative Latent Optimization models, can efficiently produce high quality reconstructions of mid-infrared spectra in ∼ 60% of the cases. We discuss how our method can be improved by using more training data, photometric bands, model parameters or by employing other generative networks.