VQPulsar: Pulsar Candidate Analysis via Deep Generative Model

VQPulsar: Pulsar Candidate Analysis via Deep Generative Model

연구 분야: Infrastructure

학회: International Conference on Advanced Data Mining and Applications

This study introduces a novel pulsar signal analysis method based on Vector Quantized Variational Autoencoder (VQVAE) and Generative Pre-trained Transformer (GPT), designed to tackle the prevalent issue of data imbalance in pulsar candidate datasets. The proposed method involves projecting pulsar signals into latent space using VQVAE for feature extraction and analysis, thereby enhancing the model’s ability to differentiate between pulsar and non-pulsar signals. Furthermore, we leverage GPT to generate synthetic pulsar samples, effectively balancing the dataset and improving the robustness and accuracy of model training. Experimental validation on the High Time Resolution Universe (HTRU) and Five-hundred-meter Aperture Spherical Telescope (FAST) datasets demonstrates significant improvements in precision, recall, and F1 score. The results confirm the efficacy of our approach in achieving high-performance pulsar identification and classification, highlighting the potential of integrating VQVAE and GPT for advanced astronomical data analysis.