PocketNN: Integer-only Training and Inference of Neural Networks via Direct Feedback Alignment and Pocket Activations in Pure C++.
Jaewoo SongFangzhen LinPublished in: CoRR (2022)
Keyphrases
- neural network
- training process
- training algorithm
- feed forward neural networks
- feedforward neural networks
- multi layer perceptron
- structured prediction
- backpropagation algorithm
- pattern recognition
- training sessions
- bayesian networks
- training set
- fuzzy logic
- genetic algorithm
- test set
- back propagation
- neural network training
- artificial neural networks
- training patterns
- motor skills
- error back propagation
- multi layer
- dynamic time warping
- multilayer perceptron
- multiple sequence alignment
- feed forward
- recurrent networks
- neural network model
- echo state networks
- activation function
- recurrent neural networks
- medical images
- supervised learning
- high dimensional
- feature space