Neuromorphic computing : From a memristive device to the learning process
Mercredi 12 juillet 2017 14:00
- Duree : 1 heure
Lieu : SPINTEC, Bât 10-05 - room 434, CEA Grenoble, 17 rue des martyrs
Orateur : Steven LEQUEUX (Spintec)
In the current context of information technology, the sequential processing carried out by classical computer architectures stumbles on problems of energy consumption. Inspired by nature, especially the brain, an alternative solution appears through Artificial Neural Networks (ANN). This gives rise to a field of research that is expanding more and more called neuromorphic computing. The main objective of these ANNs is to mimic one of the most interesting abilities of the brain : its aptitude to learn. Nowadays, the evolution of knowledge and technology allows us to achieve nano-components, called memristor, which can mimic the plasticity of a biological synapse [1,2]. This achievement is a first step in the complicated path toward the realization of densely interconnected artificial neural networks.
This presentation will describe the following step in the achievement of a memristive device, how to implement a memristor within a small network in order to make it able to learn and execute tasks of patterns recognition or classification [2,3].
[1] S. Lequeux et al., Sci. Rep. 31510 (2016). [2] S. Boyn et al., Nature Com. 14736 (2017). [3] D. Querlioz et al., IEEE 103 (2015).
Contact : olivier.boulle@cea.fr
Prévenir un ami par email
Télécharger dans mon agenda