The Super-Kamiokande experiment (SK) is the water Cherenkov detector which discovered the oscillation of atmospheric neutrinos. The dominant effect of the oscillation of muon neutrinos (νμ) is the appearance of tau neutrinos (ντ). Direct detection of ντ in the atmospheric neutrino flux provides an unambiguous confirmation of neutrino oscillations. SK uses machine learning techniques of neural networks to segregate ντ charged-current interactions from the interactions of the atmospheric muon and electron neutrinos (νe), with which in 2018, it excluded the hypothesis of no ντ appearance with a significance level of 4.6σ. The sub-dominant νμ oscillation mode, which is the change of νμ to νe, is studied at SK to determine mass hierarchy. Currently, ντ interactions form the biggest background to the mass hierarchy signal in the SK analysis. This poster will discuss improvements in the ντ identification algorithm and discuss corresponding improvements in the search for tau neutrinos and the suppression of mass heirarchy backgrounds.