The machine learning has evolved significantly in recent years, enabling the language models (LLMs) and computer vision can perform complex tasks with great precision.
However, many supervised learning techniques rely on large labelled datasets, which poses a limitation in terms of scalability and applicability to new domains.
In this context, the Zero-learningShot (ZSL) This is a strategy that enables models to make predictions about classes or tasks for which they have not been explicitly trained.
Read on to find out how this type of learning is implemented in the training of artificial intelligence models.
Zero-Shot Learning: Learning without the need for labelled data
The Zero-learningShot is a technique or strategy within artificial intelligence that enables a model to make inferences about data that has never seen during training.
Unlike the traditional approach, which relies on large volumes of labelled training data, ZSL enables models to spread their knowledge to new concepts without requiring specific prior samples.
This is made possible by advanced semantic representations, which enable language and vision models to link new information to previously acquired knowledge.
For example, if a model has been previously trained to recognise images of dogs and cats, but has never seen an image of a wolf, ZSL may enable it to classify the latter correctly based on textual descriptions or attributes shared with dogs and cats.
How do language models learn (LLMs) to generalise?
Share in:
Español 
English (UK) 
Français