Neural Network Action Policy Verification via Predicate Abstraction.
Marcel VinzentMarcel SteinmetzJörg HoffmannPublished in: ICAPS (2022)
Keyphrases
- neural network
- action selection
- high level
- artificial neural networks
- neural network model
- hardware designs
- joint action
- back propagation
- action space
- optimal policy
- bounded model checking
- pattern recognition
- state action
- neural nets
- verification method
- genetic algorithm
- decision theoretic planning
- radial basis function
- model checking
- first order logic
- neural network is trained
- feed forward neural networks
- expected reward
- fuzzy artmap
- partially observable domains
- function approximators
- face verification
- network architecture
- prediction model
- multilayer perceptron
- reasoning about actions
- human actions
- self organizing maps
- fault diagnosis
- state space
- dynamic programming
- reinforcement learning