Summary
Building and training a CNN will only succeed with hard work, choosing the model, the right datasets, and hyperparameters. The model must contain convolutions, pooling, flattening, dense layers, activation functions, and optimizing parameters (weights and biases) to form solid building blocks to train and use a model.
Training a CNN to solve a real-life problem can help sell AI to a manager or a sales prospect. In this case, using the model to help a food-processing factory solve a conveyor belt productivity problem takes AI a step further into everyday corporate life.
A CNN that recognizes abstract concepts within an image takes deep learning one step closer to powerful machine thinking. A machine that can detect objects in an image and extract concepts from the results represents the true final level of AI.
Once the training is over, saving the model provides a practical way to use it by loading it and applying it to new images to classify them. This chapter...
