Encouraging orthogonality between weight vectors in pretrained deep neural networks.
Karol GrzegorczykMarcin KurdzielPiotr Iwo WójcikPublished in: Neurocomputing (2016)
Keyphrases
- weight vector
- neural network
- input vector
- learning rate
- linear combination
- pattern recognition
- hyperplane
- objective function
- artificial neural networks
- loss function
- multilayer perceptron
- neural network model
- data sets
- feature values
- support vector machine
- back propagation
- genetic algorithm
- principal components
- mean shift
- learning algorithm
- data points
- multi objective
- evolutionary algorithm
- reinforcement learning