In this talk I will discuss two recent projects: (i) latent space anomaly detection with Variational AutoEncoders (VAEs), (ii) the contrastive learning of jet representations (JetCLR). In the first part I will present results on boosted top/QCD jet anomaly detection with VAEs using a range of latent space structures. While in the second part I will discuss how we can use an unsupervised machine learning tool called Contrastive Learning of Representations (CLR) to define a representation of our jet data which is (approximately) invariant to the relevant symmetries in the problem and which is discriminative within the dataset it is trained on.