INTRODUCTION
In machine learning, machines become smarter when they are provided with properly labelled data set during training period. What happens when machines are attempted to classify instances whose classes were not seen in training period ? This led to the rise of new learning method called Zero-shot Learning where instances can be classified even though the classes were not seen before.
PROCESS OF ZERO-SHOT LEARNING
In the training stage, knowledge about the attributes is captured, and in the inference stage, this knowledge is used to categorize instances among a new set of classes.
HOW ZERO-SHOT LEARNING WORKS ?
Let us consider an example, a child would have no problem recognising a leopard if it has seen a cheetah before and read somewhere that a leopard looks similar to a cheetah, but has rosette markings instead of solid spots.
In case of machines, the Zero-Shot Learning recognition relies on the existence of a labelled training set of seen classes and the knowledge about how each unseen class is semantically related to the seen classes. This knowledge transfers from seen classes to unseen classes occurs through a high dimensional vector space called semantic space.
Likewise here is an another example how Zero-shot learning works. It extracts (knowledge) attribute features from seen classes such as tiger, horse, panda during training period and predicts zebra which is an unseen class with the help of attribute features (knowledge) from seen classes.
WHY DO WE NEED ZERO-SHOT LEARNING ?
Rise in domain categories causes difficulty in collecting data of all categories. Zero-shot learning aims at predicting a large number of unseen classes using only labelled data from a small set of classes and external knowledge about class relations. Hence unseen classes can also be predicted without the use of large amount of data from all categories.
Zero-shot Learning method is used where ML models classifies data with less or even no labelled examples, thus leveraging supervised learning with no additional training data.
CONCLUSION
Zero-Shot Learning can reduce training and help systems be accurate when confronted with unexpected data. It can be implemented in many applications where there is inaccurate results due to lack of data.
Thanks for reading. I hope it is helpful.
