Implicit Unlikelihood Training: Improving Neural Text Generation with Reinforcement Learning.
Evgeny LagutinDaniil GavrilovPavel KalaidinPublished in: EACL (2021)
Keyphrases
- text generation
- reinforcement learning
- natural language generation
- supervised learning
- network architecture
- neural network
- training set
- natural language
- fitted q iteration
- recurrent networks
- machine learning
- reinforcement learning algorithms
- markov decision processes
- training process
- training algorithm
- state space
- learning process
- learning algorithm
- recurrent neural networks
- optimal control
- training samples
- learning rules
- multi agent
- information retrieval