LinSyn: Synthesizing Tight Linear Bounds for Arbitrary Neural Network Activation Functions.
Brandon PaulsenChao WangPublished in: CoRR (2022)
Keyphrases
- activation function
- neural network
- lower bound
- upper bound
- back propagation
- feed forward neural networks
- artificial neural networks
- hidden layer
- hidden neurons
- neural architecture
- connection weights
- feed forward
- nonlinear functions
- neural nets
- learning rate
- multilayer perceptron
- feedforward neural networks
- network architecture
- basis functions
- sigmoid function
- radial basis function
- training phase
- pattern recognition
- multi layer
- multi layer perceptron
- fuzzy neural network
- learning algorithm
- autoregressive
- input space
- fuzzy logic
- feature vectors