Aetherscan

Aetherscan is a deep learning model designed for anomaly detection in radio astronomy data, specifically within the context of the Search for Extraterrestrial Intelligence (SETI). The model implements a two-stage architecture as part of the Breakthrough Listen deep-learning SETI pipeline. In this setup, a Beta-VAE encoder first compresses each observation from a six-observation cadence—structured as three ON and three OFF observations (ABACAD)—into an eight-dimensional latent representation. These latent vectors are then concatenated, and a Random Forest classifier evaluates the sequence to determine whether it should be classified as a technosignature candidate. The model is compatible with Keras, and can be loaded using various backend options, including JAX, Torch, and TensorFlow. Instructions are provided for integrating the model with Keras, and it can also be used in environments such as Google Colab and Kaggle. The repository hosts the released model weights, allowing users to access and deploy the model for research or analysis purposes. Aetherscan is distributed under the BSD 3-Clause license. The evidence positions it as a tool for researchers and practitioners engaged in radio-astronomy anomaly detection, particularly those working with SETI data and interested in identifying potential technosignatures. The model and its documentation are available via the Hugging Face platform.