Reinforcement learning (RL) is area of machine learning it is also known as a semi-supervised learning model(behavioral learning )in machine learning. The system isn’t trained with the sample data set. The system learns from trial and error to come up with a solution to the problem.it is employed by various software and machines to find the best possible behavior or path it should take in specific situation .
Reinforcement learning is no doubt a cutting-edge technology that has the potential to transform our world. However, it need not be used in every case. Nevertheless, reinforcement learning seems to be the most likely way to make a machine creative — as seeking new, innovative ways to perform its tasks is in fact creativity.
RL is usually modeled as a Markov Decision Process (MDP).
Eg : Take the example of the need to train a ROBOT to a set of stairs . The ROBOT changes its approach to navigating the terrain based on the outcome of its actions. When the ROBOT falls the data is recalibrated so the steps are navigated differently untill the ROBOT is trained by trial and error to understand how to count stairs .
However, the problems we face in the real world can be extremely complicated in many different ways and therefore a typical RL algorithm has no clue to solve. For example, the state space is very large in the game of GO, environment cannot be fully observed in Poker game and there are lots of agents interact with each other in the real world. Researchers have invented methods to solve some of the problems by using deep neural network to model the desired policies, value functions or even the transition models, which therefore is called Deep Reinforcement Learning. This article makes no distinction between RL and Deep RL.
Main points in Reinforcement learning (RL)
- Input : The input will be at initial state from which the model will start . - Output : Their are many possible outcome as their are variety of solution to a particular problem . - Training : The training is based on the input .The model will return a state and the user will decide to reward and punish the model based on its output . - The model keeps continues to learn. - The best solution is decided based on maximum reward.
Types of Reinforcement learning (RL)
Their are two types of Reinforcement learning (RL)-
1. Positive Reinforcement learning -
Positive reinforcement is defined as when an event , occurs due to particular behavior , increases the strength or the frequency of behavior. In other words it has a positive effect on the behavior .
Advantages of Positive Reinforcement learning are :
- Maximizes Performance . - Sustain change for a long period of time .
Disadvantages of Positive Reinforcement learning are :
- Too much reinforcement can lead to overload of states which can diminish the results
2 . Negative Reinforcement learning -
Negative reinforcement is defined as the strengthening of a behavior`because a negative condition is stopped or avoided .
Advantages of Negative Reinforcement learning are :
- Increase behavior - Provide defiance to minimum standard of performance.
Disadvantages of Negative Reinforcement learning are :
- It only provide enough to meet up the minimum behavior .
Practical Applications of Reinforcement Learning :
- Reinforcement learning can be used in robotics for industrial automation .
- Reinforcement learning can be used in machine learning and data processing .
- Reinforcement learning can be used games .
- Reinforcement learning can be used traffic light control .
Traffic light control
Reinforcement Learning can be used in large environment in following situations
- A model of environment is known but an analytical solution is not available . - Only a simulation model of the environment is given ( the subject of simulation based optimization.) - The only way to collect information about the environment is to interact with it.
Intuitions from other disciplines
RL has a very close relationship with psychology, biology and neuroscience. If you think about it, what a RL agent does is just trial-and-error: it learns how good or bad its actions are based on the rewards it receives from the environment. And this is exactly how human learns to make a decision. Besides, the exploration and exploitation problem, credit assignment problem, attempts to model the environment are also something we face in our everyday life.
The Economics theory can also shed some light on RL. In particular, the analysis of multi-agent reinforcement learning (MARL) can be understood from the perspectives of game theory, which is a research area developed by John Nash to understand the interactions of agents in a system. In addition to game theory, MARL, Partially Observable Markov Decision Process (POMDP) could also be useful to understand other economic topics like market structure (e.g.monopoly, oligopoly, etc), externality and information asymmetry.
Is reinforcement learning the future of machine learning?
RL still has lots of problems and cannot be used easily. Yet, as long as more efforts are put in solving the problems, RL would be influential and impactful in the following ways:
Assisting human: Maybe it is too much to say RL can one day evolve into artificial general intelligence (AGI), but RL surely has the potential to assist and work with human. Just imagine a robot or a virtual assistant working with you and taking your actions into its considerations to take actions in order to achieve a common goal. Wouldn’t it be great?
Understanding the consequences of different strategies: Life is so amazing because time will not go back and things just happen once. Yet, sometimes we would like to know how things could be different (at least in the short term) if I took a different action? Or would Croatia has a greater chance to win the 2018 World Cup if the coach used another strategy? Of course, to achieve this we would need to model the environment, transition functions and so on perfectly and also analyse the interactions between the agents, which seems to be impossible at the moment.
Sometimes machine learning is only supporting a process being performed in another way, for example by seeking a way to optimize speed or efficiency.
When a machine has to deal with unstructured and unsorted data, or with various types of data, neural networks can be very useful.
Thus, reinforcement learning has the potential to be a groundbreaking technology and the next step in AI development.
What I have presented here are the insights of What is Reinforcement Learning and Practical Applications. I hope you learned something today.
Always remember that solid business questions, clean and well-distributed data always beat fancy models.
Feel free to leave a message if you have any feedback, and share with anyone that might find this useful