Self-supervised clustering on image-subtracted data with deep-embedded self-organizing map

ABSTRACT Developing an effective automatic classifier to separate genuine sources from artifacts is essential for transient follow-ups in wide-field optical surveys. The identification of detections the subtraction after image differencing process a key step such classifiers, known as real-bogus classification problem. We apply self-supervised machine learning model, deep-embedded self-organizing map (DESOM) this ‘real-bogus’ DESOM combines autoencoder and perform clustering order distinguish between real bogus detections, based on their dimensionality-reduced representations. use 32 × normalized detection thumbnails input DESOM. demonstrate different model training approaches, find that our best shows missed rate $6.6{{\ \rm per\,cent}}$ with false-positive $1.5{{\ per\,cent}}$. offers more nuanced way fine-tune decision boundary identifying likely when used combination other types e.g. built neural networks or trees. also discuss potential usages its limitations.