KO
|
EN
gitlite — search
Search
#python
#react
#python3
#javascript
#go
#nodejs
#java
#machine-learning
#vue
#tutorial
#php
#express
GENE
★ 24
Open GitHub ↗
Generative Exploration and Exploitation
Download README (.md)
Explore Similar Repositories
TwoWaysToImproveCSC
:
This is the official code for paper titled "Exploration and Exploitation: Two Ways to Improve Chinese Spelling Correction Models".
self-imitation-learning-pytorch
:
This is the pytorch implementation of ICML 2018 paper - Self-Imitation Learning.
advanced-bandit-problems
:
More about the exploration-exploitation tradeoff with harder bandits
sample-efficient-bayesian-rl
:
Source for the sample efficient tabular RL submission to the 2019 NIPS workshop on Biological and Artificial RL
// repository documentation
Was this content helpful?
★ 0
(0 ratings)
Select Rating:
★
★
★
★
★
Submit Feedback
Recent Feedback
×
Download README
Do you want to download the
README.md
file for
GENE
?
Download (.md)
## GENE This is an improved version of [GENE](https://z0ngqing.github.io/paper/aaai-jiechuan20.pdf). In the original version, the states are generated from the decoder of VAE. We have to check whether the gerenated states are safe or not, which still requires the prior knowledge about the environment. In this version, we select initial states from the history states in memory buffer according to the estimated densities. Also, we replace the rejection sampling with rank-based sampling in PER, which is more practical. When alpha is set to be 0, GENE degrades to History. <img src="maze1.png" alt="maze" width="300"> We test GENE in the sparse-reward maze task, where the agent learns to reach the landmark. The results are shown in the Figure 1. <img src="success_rate.png" alt="maze" width="400"> The distribution of generated states during the training: <img src="generated_states.gif" alt="s" width="400"> <img src="generated_states_1.gif" alt="s1" width="400"> ## Citation If you are using the codes, please cite our paper. @inproceedings{jiang2020generative, title={Generative Exploration and Exploitation}, author={Jiang, Jiechuan and Lu, Zongqing}, booktitle={Proceedings of the AAAI Conference on Artificial Intelligence}, volume={34}, number={04}, pages={4337--4344}, year={2020} }
