Google AutoML, Possible to know Model Structure?

Question

I am a Free Trial user of GCP and studying image classification service with AutoML Vision. And I got a basic question about this service. We know that there are many Object Recognition Models and Object Detection Models. It can be RCNN, YOLO, etc.

I used AutoML Vision with my photo datas and I got my custom model. There was no way to know about the structure of my Model. For example, which 'models' are used to train my datas, which 'network' is used, what are the 'parameters' that are set for my model, and way of 'tunning' of my model. And I want to know if there is any way to see those informations about my custom model that is created from AutoML Vision. If it's trade secret, I know that I need to give up about those information.

Sum : I want to know about 'Model', 'Network', Parameters', 'Way of Tunning' of my custom. (What kind of values are used for my model)

Answer 1

AutoML is an Automated Machine Learning product (this is, AI creating AI). The insights you would like to have are just not possible because the model is much more complex than "a CNN" or "a LSTM". AutoML uses transfer learning to combine pretrained Google models with further operations so that you get really accurate results with little training and no programming.

The idea of AutoML is precisely to allow people outside data science to build neural network models (or allow machine learning engineers to have some first insights on their data potential).

You can anyway try to analyze the saved model with a script like this in order to get some insights on the ops that take place. I did it myself with an image classification AutoML model and got this: