CodePudding user response:
From research to application of neural network algorithm usually has two links, the first step is to design and training model; Second link model is deployed on products, inference is usually considered to be deployed to product implementation framework needed after operation, a model of a given input data from a data, get a result similar to the reasoning, but the first link in the above said, actually also involve the reasoning, reasoning only training is only a fraction of them, the general training + + loss function data sets including reasoning, to execute the reasoning process of the software package can be thought of as a reasoning framework,