AI Detects Gamma-Ray Transients with Fermi-LAT

This paper presents a framework for detecting transient gamma-ray phenomena using a self-supervised Convolutional Long Short-Term Memory (ConvLSTM) network. The model is trained on simulated Fermi-LAT sky data to reconstruct expected emission. Departures from the learned baseline are quantified using pixel-wise mean-squared residual maps. Statistically motivated anomaly criteria and spatial coherence filtering are used to flag localized, time-dependent excesses, potentially indicating high-variable sources or transient events. The resulting pipeline is deployed on Fermi-LAT daily maps, providing a benchmark for evaluating anomaly-detection strategies on long-duration datasets. The system can be affected by instrumental non-stationarities. The use of alphaXiv, CatalyzeX, DagsHub, Gotit.pub, Hugging Face, and ScienceCast indicates a commitment to open science and reproducible research.

_Context: This intelligence report was compiled by the DailyOrbitalWire Strategy Engine. Verified for Art. 50 Compliance._