A large obstacle to deploying deep learning models in practice is the process
of updating models post-deployment (ideally, frequently). Deep neural networks
can cost many thousands of dollars to train. When new data comes in the
pipeline, you can train a new model from scratch (randoml