Advances in ML for Higgs physics in the ATLAS experiment

Machine learning plays a core role in the Higgs physics program at the ATLAS experiment, particularly in analyses of hadronic final states. We will present an overview of some of the novel methods used in recent run 2 results in the Higgs sector, with a focus on H(bb) and H(cc). We will also discuss advances in jet flavor identification using permutation-invariant Transformer-based neural networks in both the resolved and merged jet regimes. These models include physics inspired auxiliary tasks, such as the reconstruction of secondary vertices. In addition to enhancing model interpretability, the auxiliary tasks also improve flavor identification performance, allowing for higher precision probes of hadronic Higgs decays for run 3.