Exploring different architectures of NSAI
Although NSAI is still a relatively niche and emerging field of study, researchers from the MIT-IBM collaboration and Google’s DeepMind have already contributed interesting research concerning NSAI architectures. In this section of the chapter, we will further explore and discuss some of these main NSAI architectures that have been proven effective.
Neuro-Symbolic Concept Learner
The main objective of the Neuro-Symbolic Concept Learner (NSCL) architecture is to produce a model capable of learning to identify objects in an image and parsing and understanding their semantics and linking relationships [3]. NSCL is based on the concept that humans can understand visual concepts through their ability to bridge between vision and language. For example, let us assume we are shown a photo of a blue elephant. We immediately identify that the “object” captured in the picture is an elephant. We also understand that the color...
