Transfer learning across heterogeneous tasks using behavioural genetic principles

We explore the use of Artificial Neural Networks (ANNs) as computational models capable of sharing, retaining and reusing knowledge when they are combined via Behavioural Genetic principles. In behavioural genetics, the performance and the variability in performance (in case of population studies) stems from structure (intrinsic factors or genes) and environment (training dataset). We simulate the effects of genetic influences via variations in the neuro-computational parameters of the ANNs, and the effects of environmental influences via a filter applied to the training set. Our approach uses the twin method to disentangle genetic and environmental influences on performance, capturing transfer effects via changes to the heritability measure. Our model captures the wide range of variability exhibited by population members as they are trained on five different tasks. Preliminary experiments produced encouraging results as to the utility of this method. Results provide a foundation for future work in using a computational framework to capture population-level variability, optimising performance on multiple tasks, and establishing a relationship between selective pressure on cognitive skills and the change in the heritability of these skills across generations.