Hi sir,
Is there any possibility to use any reinforcement algorithms for feature selection procedure.
Or will it be used for prediction process? Since it is based on action and reward based process, will it take longer time to execute? can you please discuss the merits of reinforcement learning when compared to supervised machine learning?
Thanks,
Thenmozhi
I would recommend this paper for the fundamentals: Preprint Model-Based Reinforcement Learning for Atari
RL can be used to train an intelligent agent to for example play a game like Atari.
Reinforcement learning is one of three basic machine learning paradigms, alongside supervised learning and unsupervised learning.
I have not heard of any use cases in feature selection. RL does take a lot of compute to train, yes.
Cheers,
Raoul
Thenmozhi Thangaraj How about you look at below
Conference Paper Feature Selection for Reinforcement Learning: Evaluating Imp...
Conference Paper Embedded Incremental Feature Selection for Reinforcement Learning.
I would say, yes it is possible to use Reinforcement Learning for selecting features. You can take a look at the method followed in this paper.
The selection of a machine learning algorithm depends on your goal and the state of your system. If you have a continuous system and you want to train a network to predict something in run time through interacting with the environment and get feedback from the environment to retrain the network then you might use RL.
Conference Paper Q-MIND: Defeating Stealthy DoS Attacks in SDN with a Machine...
Thanks a lot for your valuable suggestions Syed sir, Stanley sir, Raoul sir. I will look into the materials.
Thanks,
Thenmozhi
Reinforcement Learning (RL) is about an agent that learns to act in an environment. He learns by getting rewards with respect for his actions. The agent's goal is to maximize the reward. The main applications of Reinforcement learning are Games and Robotics. I utilize RL to trade stocks as trading stocks is like playing a game where you need to maximize reward.
I can not think of an agent that will learn to make feature selection.
Feature selection is done mostly by measuring the correlation between the features and the target (for supervised learning). The correlation between the feature themself also interesting as we do not want too much similar features.
There are other methods for feature selection or optimization e.g PCA - principal component analysis. Lastly, feature selection can be done with some decision tree algorithms with the scikit-learn library, python, where the algo can produce the importance of each feature.
Hi... The following article may be useful for you..
Chapter Cyber Secure Man-in-the-Middle Attack Intrusion Detection Us...
with regards
Jayapandian N
Check out the following article, it might be useful for you:
https://link.springer.com/chapter/10.1007%2F978-3-030-20518-8_65
- Badges
- Science topic