Mixture of experts
Mezcla de expertos
- Date
- 1991 · confidence medium
- Place
- Cambridge, Massachusetts, United States
- Credited to
- Steven J. Nowlan, Michael I. Jordan, Robert A. Jacobs y Geoffrey Hinton
- Field
- Knowledge & symbols
What it is
A machine-learning architecture that divides a problem among several specialized "expert" subnetworks, coordinated by a gating network that decides which to consult for each input, proposed by Robert Jacobs, Michael Jordan, Steven Nowlan and Geoffrey Hinton.
What it led to
- an early form of ensemble learning in neural networks
- conceptual basis of today's sparse mixture-of-experts language models
- boost to modular, scalable deep-learning architectures