AI in the Cosmos

TL;DR

The paper addresses the explosion of data in astronomy and the need for AI/ML and generative AI to unlock maximal information from surveys. It surveys ML/AI applications in astrophysics, demonstrates a CNN-based surrogate for blazar SED modeling trained on SOPRANO-generated spectra, and discusses the integration of generative AI tools within a Human-Guided AI (HG-AI) framework to maintain interpretability and ethics. Key findings include a classification study where LightGBM achieved $88\%$ recall/precision in BCUs and a surrogate CNN that reduces SED evaluations to milliseconds, enabling real-time parameter inference. The work argues that HG-AI and astroLLM-based workflows can enhance discovery potential while ensuring transparency, reproducibility, and responsible AI use in astrophysics.

Abstract

Artificial intelligence (AI) is revolutionizing research by enabling the efficient analysis of large datasets and the discovery of hidden patterns. In astrophysics, AI has become essential, transforming the classification of celestial sources, data modeling, and the interpretation of observations. In this review, I highlight examples of AI applications in astrophysics, including source classification, spectral energy distribution modeling, and discuss the advancements achievable through generative AI. However, the use of AI introduces challenges, including biases, errors, and the "black box" nature of AI models, which must be resolved before their application. These issues can be addressed through the concept of Human-Guided AI (HG-AI), which integrates human expertise and domain-specific knowledge into AI applications. This approach aims to ensure that AI is applied in a robust, interpretable, and ethical manner, leading to deeper insights and fostering scientific excellence.

Figures (4)

• Figure 1: Number of astrophysics articles containing the term "machine learning" over the years.
• Figure 2: Likelihood distributions for sources in the test sample (left panel) and BCUs (right panel), showing probabilities of being classified as BL Lac or FSRQ. Adapted from Ref. [7].
• Figure 3: Workflow of the method for training a CNN for blazar SED modeling. A subset of parameters is selected from the entire parameter space using Latin hypercube sampling and passed to SOPRANO to compute the corresponding SEDs. Both the entire parameter range and the selected subset, along with the generated SEDs, are then passed to the CNN, which learns and predicts the relationships between all parameters and SEDs. This network combined to MultiNest can fit the observed data.
• Figure 4: The broadband SEDs of Mrk 421 and CTA 102 are modeled under the SSC and EIC scenarios, respectively. The plots are adapted from Refs. 10 and 11.