/    /  Machine Learning- Reinforcement Learning: Introduction

Reinforcement Learning: Introduction

 

Reinforcement learning is a method for teaching an autonomous agent that observes and acts in its surroundings to pick the best actions to achieve its objectives. In this blog, we’ll discuss reinforcement learning.

 

Reinforcement Learning is one of the exciting newer fields to emerge from Deep Learning. The idea is to get machines to learn without being told “exactly” what to do. They observe the world around them and continuously learn from experience. The idea is to find solutions using trial and error.

 

Learning to manage a mobile robot, learning to optimize factory processes, and learning to play board games are all examples of this broad topic.

 

A trainer may deliver a reward or penalty each time the agent performs an action in its surroundings to reflect the attractiveness of the ensuing state.

 

When teaching an agent to play a game, for example, the trainer could give a positive reward when the game is won, a negative reward when the game is lost, and no reward in all other situations.

 

The agent’s job is to learn from this indirect, delayed reward and select action sequences that yield the highest cumulative reward.

 

Dynamic programming methods, which are commonly employed to tackle optimization issues, are connected to reinforcement learning algorithms.

 

The robot, or agent, has a collection of sensors that allow it to monitor the state of its surroundings and a set of actions that allow it to change that state.

 

A mobile robot could have sensors like a camera and sonars, as well as commands like “go ahead” and “turn.”

 

Its job is to figure up a control plan, or policy, for selecting activities that will help it achieve its objectives. When the robot’s battery level is low, it may have the purpose of docking onto its power charger.

 

A reward function that provides a numerical value—an instant payoff—to each individual action the agent may take from each distinct condition might specify the agent’s aims.

 

The aim of docking to the battery charger can be achieved by providing a positive reward (e.g., +loo) to state-action transitions that result in an instant connection to the charger, and a reward of zero to all other state-action transitions. 

 

This reward function may be integrated inside the robot, or it may be known only to an external teacher who assigns a reward value to each robot activity. The robot’s job is to follow a set of instructions, examine the results, and learn a control strategy.

 

Manufacturing optimization challenges in which a sequence of manufacturing operations must be determined, with the reward being the value of the items produced minus the expenses associated.

 

Choosing which taxis to send for passengers in a big metropolis, where the reward to be maximized is a function of the passengers’ wait time and the taxi fleet’s total fuel expenses, is an example of sequential scheduling issues.

 

Any form of agent that must learn to pick actions that change the state of its surroundings and where the quality of each particular action sequence is defined by a cumulative reward function

 

Include scenarios in which the agent has or does not have previous knowledge of the repercussions of its actions on the environment, and scenarios in which the agent’s actions have deterministic or non-deterministic results.

 

The agent resides in a world that may be characterized by a collection of states. S. It has the ability to do any of a number of different actions A. The agent receives a real-valued reward ri each time it executes an action ai in some state st, indicating the immediate value of this state-activity transition. 

 

This results in a series of states (si), acts (ai), and instantaneous rewards (ri).

 

The agent’s job is to figure out a control strategy (S -> A) that maximizes the predicted total of these rewards.

 

The control strategy we want is one that, given any starting state, chooses behaviours that maximise the agent’s reward over time.

 

 

Here are some of the most important aspects of reinforcement learning-

 

  • Input: The input should be a starting state for the model to work from.
  • There are several alternative outputs, just as there are numerous solutions to a given issue.
  • Training: The model will return a state depending on the input, and the user will decide whether to reward or penalize the model based on its output.
  • The model is always evolving.
  • The optimal answer is determined by the highest possible payment.

 

Reinforcement may be divided into two categories: positive reinforcement and negative reinforcement.

 

  • Positive Reinforcement happens when an event occurs as a result of a certain behavior, increasing the strength and frequency of that behavior. It has proved to be very beneficial. 

 

  • Negative – It is when a behavior is strengthened as a result of a negative circumstance being avoided or halted.

 

Let’s look at some of the applications of Reinforcement Learning->

 

1. Reinforcement Learning can be used in 

a)Robotics 

b)Industrial automation 

c)Machine Learning 

d)Data Processing.

 

2. RL may be used to construct training systems that deliver students with customized education and resources based on their needs.

 

Reference

Reinforcement Learning: Introduction