Reinforcement Learning is an area of Machine Learning concerned with various software, a bunch of algorithm and programming. RL is different than Supervised, Semi-supervised and Unsupervised Learning. The RL doesn’t have input-output enabled learning datasets as we see in other Machine Learning algorithmic approaches.
Reinforcement Learning differs from other Machine Learning technique because in this technique there are no data provided to the machine to attain the desired value or to get the desired result. It learned from its experiences and real-time environmental situations. As mentioned in the name itself, it is a technique in which machine reinforced or improved itself by previous experiences. A system interacts with a dynamic environment in which it must perform certain goals. In this algorithmic approach, feedback is given to the machine to perform the task. Here feedback can add or remove the data to increase the performance of the machine. The best-known example of machine learning is the navigation system of driverless cars.
Due to no dataset provided to the machine in case of Reinforcement learning techniques. Then in this case machine is trained in the real-world environmental situation to perform such task and take the decision without explicitly design to perform that specific tasks.
RL Framework
- Learn from close interaction
- Stochastic Environment
- Noisy delayed scalar evaluation
- Learn a policy
Maximize a measure of long-term performance
There are two types of RL (Reinforcement Learning)
Positive Reinforcement Learning
When an event or task occurred in the environment due to particular type of behavior in such a way that it increases the frequency and the strength of the behavior and impact positively on the behavior is known as Positive RL.
Advantages – Maximize Performances, Sustain change for a long period of time.
Negative Reinforcement Learning
When a behavior is strengthening due to negative condition is stopped or avoided is known as Negative RL.
Advantages – Increase behaviour, Provide defiance to a minimum standard of performance.
Well, Known Reinforcement Algorithms –
Examples/Applications of RL –
- Autonomous Agent
Robot navigation
- Game AI
Backgammon – World’s best player
Atari games from scratch
- Adaptive Control
Helicopter pilot
- Combinational Optimization
VLSI Placement
[…] Reinforcement Learning […]