Poster abstract details

Recent astronomical surveys like Gaia and GALAH provide scientists with an abundance of data, in terms of not only the number of observations, but also the number of dimensions. When visually analyzing such high dimensional datasets, it is crucial to reduce the number of dimensions to two or three, so that we can directly visually present the data for exploration purposes. Only after then we are able to visually search for underlying patterns or outliers in the data. In this poster, we propose the first steps towards a dimension reduction method that is scalable in terms of computational speed, feasible for cluster separation, and predictable for end-users. The high-dimensional data is first preprocessed using a technique called mean shift, where points within each potential cluster in the high-dimensional space move towards the center of the cluster. Then ISOMAP is used to perform dimension reduction. The proposed method is tested using both synthetic and real astronomical datasets including Gaia DR2 and GALAH’s recent data release, and its performance and power to depict separate clusters in the projection is compared with that of t-SNE, a dimension reduction method with high cluster separation.