We utilize knowledge from source models to improve learning in the target task. Apart from providing capabilities to reuse already-built models, transfer learning may assist learning the target task in the following ways:

• Improved baseline performance: When we augment the knowledge of an isolated learner (also known as an ignorant learner) with knowledge from a source model, the baseline performance might improve due to this knowledge transfer.
• Model-development time: Utilizing knowledge from a source model might also help in fully learning the target task, as compared to a target model that learns from scratch. This, in turn, results in improvements in the overall time taken to develop/learn a model.
• Improved final performance: Higher final performance might be attained by leveraging transfer learning.

Readers should note that one or more of these gains are possible, which we will discuss at length in the coming chapters. This is depicted in the following diagram, showcasing better baseline performance (higher start), efficiency gains (higher slope), and better final performance (higher asymptote):

Transfer learning has been applied and studied in the context of inductive learners, such as neural networks, and Bayesian networks. Reinforcement learning is another area where possibilities with transfer learning are being explored. Thus, the concept of transfer learning is not limited to deep learning.

In this chapter and the subsequent ones, we will be limiting our scope of utilizing transfer learning to the context of deep learning.