Highly Accurate CNN Inference Using Approximate Activation Functions over Homomorphic Encryption.
Takumi IshiyamaTakuya SuzukiHayato YamanaPublished in: IEEE BigData (2020)
Keyphrases
- highly accurate
- activation function
- homomorphic encryption
- neural network
- artificial neural networks
- feed forward
- multilayer perceptron
- hidden layer
- high accuracy
- high quality
- learning rate
- data sharing
- privacy preserving
- neural nets
- back propagation
- network architecture
- bayesian networks
- basis functions
- radial basis function
- fuzzy logic
- public key
- training phase
- relational databases
- recurrent neural networks
- decision making
- learning algorithm