Handling Overfitting and Imbalance Data in Modelling Convolutional Neural Networks for Astronomical Transient Discovery

Given development in telescope and image acquisition technology, efficient processing and timely discovery of events have become challenges for astronomers and data scientists around the globe. Among modern sky survey projects, GOTO (Gravitational-wave Optical Transient Observer) is searching for transient events with new breed of optical survey telescopes, which allow a faster and deeper assessment. As compared to conventional machine learning techniques that have been sup-optimal to identify new sources from a large pool of candidates, this works presents the application of convolutional neural networks, with different data augmentation methods being explored to handle issues of overfitting and imbalance data.