Tasfia Model

[jnnchi]

• Finish infrastructure to have use unlabeled training followed by labeled fine-tuning (using ResNet)
• Begin unlabeled training + labeled training
• Work on implementing some of the features from the other streams of the project, including LR scheduling, and perhaps L2 regularization
• Think about whether we want to be training on unlabeled augmented data, or only original. This will be a semi-complex question to answer. Refer to the research, if they have augmented their data.