Quantifying the Right Combination of Knowledge and Population Diversity for Transferred Populations

Transfer Learning (TL) is a process of leveraging knowledge from one problem called the source problem (S) to a related problem called the target problem (T). Experiments and previous research (see Section II) showed that the knowledge from solving a related problem and population diversity are able to aid machine learning algorithm to solve a related problem more easily. To understand the right amount of knowledge and diversity of the transferred population, we modeled the TL process using Evolutionary Computation (EC). Our model gives us more access and control over the TL components, especially the content of the transferred population. The results showed that the relationship R between the source S and the target T problems can quantify the right combination of each component of the transferred population.