Interpretations are Useful: Penalizing Explanations to Align Neural Networks with Prior Knowledge.
Laura RiegerChandan SinghW. James MurdochBin YuPublished in: ICML (2020)
Keyphrases
- neural network
- prior knowledge
- pattern recognition
- fuzzy logic
- training data
- activation function
- prior information
- incorporation of prior knowledge
- multi layer
- neural nets
- recurrent neural networks
- bayesian framework
- artificial neural networks
- fault diagnosis
- competitive learning
- fuzzy neural network
- genetic algorithm
- domain knowledge
- multilayer perceptron
- latent variables
- neural network model
- level set
- labeled data
- training process
- self organizing maps
- information retrieval
- rule extraction
- back propagation
- database