How to build good AI solutions when data is scarce data-efficient AI techniques are emerging, and that means you don't always need large volumes of labeled data to train AI systems based on neural networks
Developing AI systems based on neural networks can require large volumes of labeled training data, which can be hard to obtain in some settings. New techniques for reducing the number of labeled examples needed to build accurate models are now emerging to address this problem. These approaches encom...
| Otros Autores: | |
|---|---|
| Formato: | Libro electrónico |
| Idioma: | Inglés |
| Publicado: |
[Cambridge, Massachusetts] :
MIT Sloan Management Review
2022.
|
| Edición: | [First edition] |
| Materias: | |
| Ver en Biblioteca Universitat Ramon Llull: | https://discovery.url.edu/permalink/34CSUC_URL/1im36ta/alma991009823029006719 |
| Sumario: | Developing AI systems based on neural networks can require large volumes of labeled training data, which can be hard to obtain in some settings. New techniques for reducing the number of labeled examples needed to build accurate models are now emerging to address this problem. These approaches encompass ways to transfer models across related problems and to pretrain models with unlabeled data. They also include emerging best practices around data-centric artificial intelligence. |
|---|---|
| Notas: | "Reprint 64202." |
| Descripción Física: | 1 online resource (11 pages) : illustrations |
