Human-in-the-loop: infusing knowledge into neural networks

In this study, we propose a method for enabling humans to initialize a neural network before training. This is a novel way of involving humans in neural network training. So far, neural networks learn very well from structured data. However, human knowledge, common sense, and experiences are not necessarily easy to express as structured data. This limits the ability of neural networks to learn from humans. This study attempts to fill the gap and enhance the learning ability of neural networks. This paper explains the structure of the neural network that supports this objective and the results of the initial experiments in transferring human knowledge to neural networks.