Negative initial weights improve learning in recurrent neural networks.
Davor PavisicJean-Philippe DrayeRoberto TeranGustavo CalderonGuy CheronGaetan LibertPublished in: ESANN (1996)
Keyphrases
- recurrent neural networks
- recurrent networks
- cascade correlation
- echo state networks
- learning algorithm
- learning systems
- learning tasks
- learning process
- prior knowledge
- neural network
- reinforcement learning
- active learning
- feedforward neural networks
- input output
- reservoir computing
- hebbian learning
- neural models
- autonomous robots
- supervised learning
- control system
- genetic algorithm