Applied Deep Learning with Python

Re-Training an Old Model

With this strategy, we re-train an existing model with new data. Using this strategy, one can continuously adjust the model parameters to adapt to new phenomena. However, data used in later training periods may be significantly different from the earlier data. Such differences may cause significant changes to the model parameters, making it learn new patterns and forget old patterns. This phenomenon is generally referred to as catastrophic forgetting.

Catastrophic forgetting is a common phenomenon affecting neural networks. Deep learning researchers have been trying to tackle this problem for many years. DeepMind, a Google-owned deep learning research group from the United Kingdom, has made notable advancements in ...