Text generation method based on meta-reinforcement learning
A technology of reinforcement learning and text, applied in neural learning methods, machine learning, biological neural network models, etc., can solve problems such as poor collection of learning samples, achieve rapid adaptation to new environments, solve data shortages, and design reasonable effects
Active Publication Date: 2020-07-31
TIANJIN UNIV OF SCI & TECH
View PDF2 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
[0004] The purpose of the present invention is to overcome the deficiencies of the prior art, and propose a text generation method based on meta-reinforcement learning, which is used to solve the problem that the language generation model in the real world can quickly adapt to different scenarios for text generation and the collection of learning samples in individual scenarios is difficult. Bottleneck problem
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment Construction
[0024] The present invention will be described in further detail below in conjunction with the accompanying drawings.
[0025] Meta Reinforcement Learning (Meta RL for short) is a research direction that applies meta-learning to reinforcement learning. Its core idea is to hope that the agent can acquire sufficient prior knowledge in learning a large number of reinforcement learning tasks, and then When faced with new intensive learning tasks, it can learn faster and better, and can quickly adapt to the new learning environment.
[0026] Such as figure 1 As shown, first from the database D train Choose any task τ from i The text data of is used as the current generation environment: initialize the text generation model M, and compare the generated text with the real text as the training error Loss n , using the policy gradient method to perform a few internal gradient updates to the generative model M as M' n . Then use the updated model M' n Continue to sample the text t...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
The invention relates to a text generation method based on meta-reinforcement learning. The text generation method is technically characterized by comprising the following steps: collecting differenttypes of text data as divisions of different tasks; collecting data of a certain task randomly adopted in the text data; constructing a text generation model by adopting the recurrent neural network for processing the sequence data; generating K text tracks; performing few times of strategy gradient updating on the text generation model by utilizing the text generation track to obtain an updated text generation model; generating a new text track; respectively updating and sampling the text generation model on the plurality of tasks to obtain a representation error of a text generation track; and performing secondary gradient updating training on the original text generation model parameters until convergence. According to the method, improvement is carried out on the basis that the recurrent neural network is used for text generation in reinforcement learning, meta-reinforcement learning is used for training the intelligent agent, experience learned on multiple tasks is migrated to thetarget task, and text generation under different scenes or contexts can be rapidly achieved.
Description
technical field [0001] The invention belongs to the technical field of computer natural language processing, in particular to a text generation method based on meta reinforcement learning. Background technique [0002] Natural language processing (NLP), especially the problem of natural language generation (NLG), has long been recognized as one of the most challenging computational tasks. Natural language generation is a technology that enables computers to have the same ability to express and write as humans. It can automatically generate a high-quality natural language text after planning based on some key information and its internal expression form. From the very beginning of pattern matching generation, some simple syntax and grammatical rules were used to organize and generate text; then it was based on statistical probability models; and now with the rapid development of deep learning, natural language generation technology based on deep learning has become more Outs...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More Patent Type & Authority Applications(China)
IPC IPC(8): G06F40/205G06N3/04G06N3/08G06N20/00
CPCG06N3/08G06N20/00G06N3/044G06N3/045Y02A90/10
Inventor 赵婷婷宋亚静王嫄任德华杨巨成
Owner TIANJIN UNIV OF SCI & TECH