Deep learning in real world

Hello, I have a non-proffesional question.

I wonder how deep learning code is used in real world?
Looking at training Time which sometimes takes too long. How Deep learning code is used in real program ? Does it Train the model on every execution of “exe” file ?

Deep Learning is trained on data not exe Files. Lets take an example.
Lets say you want to make a recognition software to tell you if an image is a dog or a cat. Deep Learning Code (Model) takes thousands of images of dogs and cats. Then it will learn the difference between them using some mathematical function.

1 Like

Training time vs inference (“real world”) time have many different features.

Training time for starters is longer because we are exposing the model to considerably more samples – multiple times → a dataset set size of 10,000 samples for 10 epochs is 100,000 operations of the data being passed through the model. In addition at training time we are doing calculations on the backward pass (to train the model) – this takes times. Luckily GPUs allow us to parallelize much of this

At inference time you are only working with one (or a few samples) and the model is only being exposed to that one time – we also have the benefit of not needing to do some work on the backward pass. Depending on how big your inference batch is - oftentimes a CPU may suffice – making the real world problem much more tenable