Machine learning training methodologies for photometric redshift estimation using HSC-PDR2 Wide grizy photometry. Both deterministic (NN) and probabilistic (BNN) neural networks are trained. Top: baseline networks trained on individual datasets, either GalaxiesML (BNN-1/NN-1) or TransferZ (BNN-2/NN-2). Middle: training on a combination of both datasets (BNN-Combo/NN-Combo). Bottom: transfer-learning methodology with fine-tuning on GalaxiesML from TransferZ pretrained models (BNN-TL/NN-TL).