RNNs can generate bounded hierarchical languages with optimal memory.
John HewittMichael HahnSurya GanguliPercy LiangChristopher D. ManningPublished in: CoRR (2020)
Keyphrases
- dynamic programming
- recurrent neural networks
- optimal design
- optimal solution
- asymptotically optimal
- expressive power
- closed form
- memory usage
- automatically generate
- memory requirements
- profit maximizing
- finite state automata
- feedforward neural networks
- fuzzy neural network
- neural network
- main memory
- worst case
- information extraction
- reinforcement learning
- learning algorithm