Post by simranratry20244 on Feb 12, 2024 1:51:59 GMT -5
The term Artificial Intelligence (AI) encompasses an amalgam of technologies that respond to a common objective. If we look at the syllabus of any undergraduate subject that responds to this name, we find a very diverse salad of techniques and algorithms . Without going any further, one of the reference books – Artificial Intelligence: a modern approach , by Stuart Russell and Peter Norvig – discusses techniques as heterogeneous as search algorithms, constraint programming, game theory, logic, ontologies or planning algorithms. , among others.
One of the dominant techniques over the last decade Colombia Telemarketing Data when it comes to AI systems in production is machine learning . The popularization of generative AIs such as GPT-4, DALL-E or Stable Diffusion has contributed to this notoriety, especially since last year . These generative AIs are based on a branch of machine learning called deep learning . In this article we will see that these techniques depend largely on data and its quality, and how the concept of “Data-Centric Artificial Intelligence” comes to pay attention to them.
How does machine learning work? Within machine learning we also find an enormous variety of algorithms and architectures, such as neural networks , which are the substrate of deep learning . Beyond the notable differences, all these algorithms have in common that they require a set of data or initial examples that show, for each input to the algorithm, what the correct output is.
From these examples, the algorithms infer patterns and learn to generalize , so that when they receive new input that is not in the initial set, they can predict what the correct output is. For example, generative image models from text , such as Stable Diffusion, are built using huge volumes of data containing description-image pairs . Once trained, they are able to create an original image that responds to a new description.
This way of building software provides a great advantage, since it is easier to explain “what” task the algorithm has to do through examples than to have to formalize “how” it is done through a programming language. It is not necessary to make a very thoughtful and in-depth representation of the process that you want to automate, you simply need to give input and output examples.