What is neural structure learning in tensorflow?

Question

• What is neural structure learning?
• what is the difference between neural network vs neural structure learning?

Answer 1

It seems that you actually mean "neural structured learning".

From the Tensorflow webpage on their neural structured learning framework, it seems it is just an umbrella term to two types of regularizations: Neural Graph Learning and Adversarial Learning.

Basically you add an extra term to your training loss where you force the internal representations for certain input to be close to the representation of its "neighbors".

The neighborhood is expressed either as a graph (hence leading to neural graph learning) or as a neighbor/no neighbor relation (hence leading to adversarial training).

Therefore, neural structured learning refers to normal neural networks that are trained with extra knowledge on what inputs are "close" to each other.