Speaker
Description
Developing automated algorithms for detecting anomalies is increasingly essential for uncovering previously unknown phenomena in astrophysics and cosmology from large volumes of radio spectrograms. To achieve this, we explore machine learning techniques for anomaly detection in the time-frequency dynamic spectra of the radio data. We evaluate our algorithms on simulated SPARKESX: Single-dish PARKES data sets for finding the uneXpected, enabling us to apply them to real, unlabeled data. We begin with essential preparation for anomaly detection, including feature extraction with a supervised Convolutional Neural Network (CNN), dimensionality reduction via Principal Component Analysis (PCA), and visualisation using Uniform Manifold Approximation and Projection (UMAP). Based on the prerequisites, we utilise unsupervised learning techniques, including Isolation Forest (IForest) and Local Outlier Factor (LOF), for anomaly detection. We discuss their performance and limitations, then introduce novel approaches for anomaly detection: ensemble learning of unsupervised learning methods and active learning using Astronomaly and Protege. We expect the new approaches to be more efficient and reliable for accurately detecting anomalous astrophysical signals than previous methods.
| Apply for student award at which level: | None |
|---|---|
| Consent on use of personal information: Abstract Submission | Yes, I ACCEPT |