Brief technical note on linearizing recurrent neural networks (RNNs) before vs after the pointwise nonlinearity.
Marino PaganAdrian ValenteSrdjan OstojicCarlos D. BrodyPublished in: CoRR (2023)
Keyphrases
- recurrent neural networks
- pointwise
- weighted sum
- feed forward
- long short term memory
- neural network
- recurrent networks
- echo state networks
- artificial neural networks
- hidden layer
- feedforward neural networks
- polynomial approximation
- reservoir computing
- neural model
- nonparametric regression
- confidence intervals
- cascade correlation
- decision making
- nonlinear dynamic systems
- state space
- roc curve