Time Series with PyTorch

11

Transfer Learning and Global Modelling

“Transfer learning is probably where neural networks have a unique advantage over shallow models.” —Andriy Burkov, The Hundred-Page Machine Learning Book

We learned earlier in the book that a key aspect of machine learning models, such as neural networks, is that they learn to map the data we input relative to their outputs. In order to adapt to the data we set these models up with various hyperparameters that allow our models to capture the optimal function for data that is passed through. This process is a means of controlling model complexity to adjust functions that suit the underlying patterns in our data, although we can also conceptualize this process as a complex form of memory compression.