Browsing by Author "Mohale, Koketso"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    Enabling unsupervised discovery in astronomical images through self-supervised representations
    (Oxford University Press, 2024) Mohale, Koketso; Lochner, Michelle
    Unsupervised learning, a branch of machine learning that can operate on unlabelled data, has proven to be a powerful tool for data exploration and discovery in astronomy. As large surveys and new telescopes drive a rapid increase in data size and richness, these techniques offer the promise of discovering new classes of objects and of efficient sorting of data into similar types. However, unsupervised learning techniques generally require feature extraction to derive simple but informative representations of images. In this paper, we explore the use of self-supervised deep learning as a method of automated representation learning. We apply the algorithm Bootstrap Your Own Latent to Galaxy Zoo DECaLS images to obtain a lower dimensional representation of each galaxy, known as features. We briefly validate these features using a small supervised classification problem. We then move on to apply an automated clustering algorithm, demonstrating that this fully unsupervised approach is able to successfully group together galaxies with similar morphology. The same features prove useful for anomaly detection, where we use the framework astronomaly to search for merger candidates. While the focus of this work is on optical images, we also explore the versatility of this technique by applying the exact same approach to a small radio galaxy data set. This work aims to demonstrate that applying deep representation learning is key to unlocking the potential of unsupervised discovery in future data sets from telescopes such as the Vera C. Rubin Observatory and the Square Kilometre Array.
  • Thumbnail Image
    Item
    Unsupervised machine learning applied to radio data
    (Universty of the Western Cape, 2023) Mohale, Koketso; Lochner, Michelle
    This thesis presents work motivated by the belief that the next generation of discoveries in the field of astronomy will be made by the marriage of advanced data analysis algorithms in the form of unsupervised learning techniques, and the unprecedented volumes and complexities of data from the next generation of surveys. For several years, computers have been governed by Moore’s law, which posited that computing power would double every two years. The consequence was that computing has also become increasingly cost-effective, which has been a driving force in the ability to generate and analyse large volumes of datasets. These include machine learning advances like the use of deep learning and scalable techniques such as self-supervised learning which have been revolutionising areas of research, for example, natural language processing and computer vision. Similarly, astronomy is also met with a rapid growth in the availability of large datasets. Morden sky observing instruments such as the radio telescope MeerKAT and the optical telescope Blanco (which was used for the Dark Energy Survey) are already producing data volumes at unprecedented scales. The next generation of instruments like the Square Kilometre Array (SKA) and the Vera C. Rubin Observatory are expected to produce orders of magnitude more astronomical data at higher resolution and sensitivity. Ongoing efforts in the form of surveys and data analysis techniques in astronomy are motivated in part by outstanding questions in galaxy evolution and cosmology as well as the potential to discover new unknown phenomena.