Large water Cherenkov detectors and long-baseline neutrino experiments have been making critical contributions to our knowledge of neutrino physics, and will continue to do so in the foreseeable future. Due to the differences of neutrino source, each experiment has relatively stronger sensitivity to one or some of the neutrino oscillation parameters. For example, the atmospheric neutrinos detected at Super-Kamiokande (SK) provide sensitivity to neutrino mass ordering, whereas the accelerator neutrinos at T2K have stronger sensitivity to the CP violation phase. To achieve better measurements of these parameters, a joint oscillation analysis of SK atmospheric and T2K accelerator neutrinos is being developed. This talk introduces the analysis strategy, newly implemented features, and expected sensitivity to oscillation parameters of this joint analysis. On the other hand, the current water Cherenkov reconstruction algorithm used in SK and T2K includes several simplifying assumptions that may introduce biases to the analyses. A novel method based on a Deep Generative Neural Network is being developed to improve the reconstruction performance in water Cherenkov detectors. The construction and preliminary results of a few networks under development will also be presented.