Visualization of each anomaly finding method (LB21, left, and this work, right). Small translucent points are GZ2 galaxies, coloured by the user interest predictions of each method (red for high interest, blue for low interest). The LB21 visualization shows the initial predicted anomaly score from the Isolation Forest. Solid red points are anomalies, defined as galaxies GZ2 volunteers voted ‘Odd’. Solid black points are galaxies chosen to be rated by the user following each method, to help inform the user interest model. Our representation gathers anomalies together better (red points are more clustred in the right-hand panel) making it easier for our active learning approach to identify the part of the representation most likely to include anomalies.