A Poisson Process AutoDecoder for X-ray Sources

2025-02-03

AI Summary

X-ray astronomical observations (e.g., from Chandra and eROSITA) yield high-dynamic-range, sparse photon streams governed by Poisson processes, posing significant challenges for source detection, physical parameter inference, and anomaly detection. To address this, we introduce— for the first time—neural fields for modeling inhomogeneous Poisson processes, proposing an end-to-end framework that jointly learns continuous energy-time domain Poisson rate function reconstruction and compact representation. Leveraging an unsupervised neural auto-decoder, our method integrates Poisson likelihood loss, continuous coordinate encoding, and latent variable optimization to directly capture the statistical nature of photon arrival times. Evaluated on the Chandra Source Catalog, our approach achieves substantial improvements: significantly enhanced rate function reconstruction fidelity; 8.2% absolute gain in downstream source classification accuracy; F1-score of 0.91 for anomaly detection; and a 37% reduction in mean absolute error for physical parameter regression.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Interdisciplinary science discovery with web data mining

Abstract

X-ray observing facilities, such as the Chandra X-ray Observatory and the eROSITA, have detected millions of astronomical sources associated with high-energy phenomena. The arrival of photons as a function of time follows a Poisson process and can vary by orders-of-magnitude, presenting obstacles for common tasks such as source classification, physical property derivation, and anomaly detection. Previous work has either failed to directly capture the Poisson nature of the data or only focuses on Poisson rate function reconstruction. In this work, we present Poisson Process AutoDecoder (PPAD). PPAD is a neural field decoder that maps fixed-length latent features to continuous Poisson rate functions across energy band and time via unsupervised learning. PPAD reconstructs the rate function and yields a representation at the same time. We demonstrate the efficacy of PPAD via reconstruction, regression, classification and anomaly detection experiments using the Chandra Source Catalog.

Problem

• X-ray Observatories
• Poisson Process
• Star Signal Analysis

Innovation

• PPAD
• Poisson Process
• X-ray Signal Processing