Zero shot learning what I learn

My understanding on Zero shot learning is a process by which we can recognise or identify the raw data without being trained on it.It does not see the data from the training phase and is still able to classify it.

Let’s take an example : Computer vision is a field where computers have the ability to detect an object like humans.

ZSL consists of recognising new categories of instances without training examples, by providing a high level description of the new categories that relate them to categories previously learned by the machine.

The inspiration of the ZSL came from the human ability to detect an object just by reading the description of a object,Let’s say a man never sees a computer but he read that “it has a monitor which is like a TV and a keyboard and mouse”, so I assume that the man already knows what a TV and a typewriter is so he can figure out just from these previous experiences.

So in computer vision image classification we have training and zero-shot classes. Remember that no samples from zero-shot classes will be used during training.

The training classes model is able to figure out how to recognise the zero shot class images by it’s features like a boy and a girl class are training class and both have features of a human so if we have a human in zero shot class model easily recognise the human without being trained on it.

The Zero shot learning can be used in many fields like nlp and image classification.