Neuromorphic architectures: challenges of electronic integration of artificial neural networks ... 18 décembre 2017

Description

Neuromorphic architectures: challenges of electronic integration of artificial neural networks and case study on Kohonen models.

The design of neuromorphic architectures in state of the art technologies faces a number of challenges that we will identify. We will discuss how the solutions explored at the international level respond to them today and study the cas of the Kohonen maps implementation.

Artificial neural networks are inspired by the tremendous capabilities of the biological brain. They have constantly refined over time, regularly finding new areas of application (computer vision, machine learning, artificial intelligence, sensory prostheses ...). Several electronic substrates are currently under study and offer interesting trade-offs to reach the energy efficiency promises of natural models and the technological maturity and programming capabilities expected by the application domains. The design of neuromorphic architectures in these different technologies faces a number of challenges that we will identify. We will discuss how the solutions explored at the international level respond to them today. We will conclude with a case study of the Kohonen maps and its distributed versions that are used by LEAT to study the transposition of neuronal self-organization into digital electronics.


Sur les réseaux sociaux


Dans un CMS/LMS ou une page web standard


En tant que lien


Début de lecture

Démarrer à :