Machine learning for self adaptive systems by Faiz ul haque Zeya

Author:Faiz ul haque Zeya [Zeya, Faiz ul haque]
Language: eng
Format: azw3
Publisher: UNKNOWN
Published: 2017-05-12T04:00:00+00:00

2.12 Reinforcement learning

Reinforcement learning is about learning how and what to do things, how to plan situation to

action i.e. which action to take for a particular situation and how to maximize, i.e. reward. The

learner itself did not know which action to take as not informed by the external user, but instead

it learns. from machine learning which is the best action to take, i.e. which action will give the

best output, best reward. Reward is external output from the environment, i.e. how good is the

environment. Action does not only affect the current output but the future outputs also. Delayed

reward is a feature of reinforcement learning.

Characterizing problem is only one of the criteria of reinforcement learning. Decision

process of Markov is used in reinforcement learning. The agent senses the environment and

based on that takes the action. This is the problem of reinforcement learning. There are three

main items here in the reinforcement learning: Sense, action and goal.

There are two types of learning in machine learning. Supervised and unsupervised. In

supervised learning, learning is assisted by a teacher. Teacher presence teaches whether teacher’s

pupil is learning correctly. Unsupervised learning is learning without teacher. Back propagation

is supervised learning, as is perceptron learning. Supervised learning is not enough for

reinforcement learning. In interactive problems it is not possible to get examples of desired

behavior that can be correct and represent the situation.

Exploit and exploration tradeoff is required in reinforcement learning. Exploit is to

exploit the already known actions and exploration is to explore the new actions.

Basically reinforcement learning deals with the whole problem of dealing with the

complete problem and in uncertain condition. Machine learning research is mostly concerned

with supervised learning. Planning theories have also been developed, but without dealing with problems of developing real time solution to the problem. The point to mention here is that we

should not take sub problems but whole problems.

Reinforcement learning on the other hand works opposite, works with complete problems

of the agents who work on complete tasks. Reinforcement learning has explicit goals, actions and

can take actions that can influence the environment.

It is realized in reinforcement learning that agent will function in both types of

environment: uncertain and certain. Information handling is the task of reinforcement learning

agents. Lots of algorithms about planning in real world agent have to choose action between

many other environments real time environment.

Planning and real time action range are to be considered for reinforcement learning.

Action selection is used for selecting correct action for a task and it is used in autonomous

character. When capabilities are critical or not, is determined by reinforcement learning.

Learning is required in main areas of this field and reinforcement learning is mainly studied in

subproblems.

Subproblem plays an important role in this area and this is done by learning agents task oriented

behavior.

Reinforcement learning is a major area of machine learning. Machine learning was

considered a separate field than artificial intelligence initially. AI field was mainly dominated by

logic, symbol, and propositional and first order logic. Propositional logic and first order logic

was considered the main areas of artificial intelligence and programs like checkers and chess

were built by logicians which comprised of prolog and lisp code.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.