Machine learning is probably one of the fastest growing trends in the current decade. No matter what the product is, everyone wants to integrate machine learning capabilities in their product, we know it’s a real head-turner. So, what really is Machine Learning (ML) in simple terms?
Machine Learning is an application of artificial intelligence where a computer program mimics human understanding or cognition. Simple right? Not really. How does a computer program mimic human cognition exactly? Well, there are three techniques through which this is achieved – supervised learning, unsupervised learning, and reinforcement learning.
What is Supervised Learning?
Imagine a classroom, where a teacher has given the students a few math problems to solve. The teacher already knows the solutions to all the math problems. Here, we can say that the teacher is supervising how each student is solving the math problems, checking to see whether their solutions are correct or not. That’s exactly what supervised learning is, where the teacher is the computer programmer, the math problems are the dataset, and the students are the machine learning algorithms. The programmer works on the labeled dataset and trains the machine learning algorithms until expected accuracy is achieved. No rocket science there.
What is Unsupervised Learning?
Then, we have unsupervised learning . The difference here is that the dataset we work with is not labeled. In reference to the classroom example, in this situation, the teacher doesn’t know the solutions to the math problems. So, what to do now? Imagine another scenario, where we have a dataset of animal photos and we have to train a machine learning algorithm to find out whether the photo is of a bird or a dog, except we don’t know how these animals or birds look like. In such scenarios, first a grouping of the data is performed and then a comparison is made by the machine learning algorithm to predict the output. When the photos present certain specific information, the algorithm can use that information to distinguish between the animals successfully. For example, if the photo suggests that the animal has feathers, the algorithm predicts that the animal is a bird. But if the photo suggests that the animal is fluffy and has a curly tail, the algorithm predicts that the animal is a dog.
What is Reinforcement Learning?
Lastly, there’s reinforcement learning. In this method, the machine learning algorithms learn to react to the environment they are exposed to, on their own. In this case, there is an agent and the environment it is exposed to has a start state and end state for the agent. Imagine a scenario, where a child is learning to walk. The child tries to balance the body and move towards the parents who are showing the child’s favorite toy from a distance. The closer the child gets to the parents, higher the probability that the child will get to play with the toy. Now, from a machine learning perspective, the child is the agent and is manipulating the environment by taking steps closer to the toy. The child (agent), starts from a distance (start state) and gets rewarded only when the child is closer to the parents (end state). Otherwise, the child or the agent are not rewarded. This is how reward-based reinforcement learning works.
Product Engineer
Nidhi was one of the first employees at HighIQ. Over her tenure so far, she has successfully completed over 10+ POCs and was a key contributor in the development of Oscar (intelligent AP automation). She is a Full-Stack and Python developer with strong knowledge in Machine Learning. She also has expertise in other technologies such as C#, Mysql, PHP, Java, Javascript, AWS, Microsoft Azure, and more. Despite being one of the youngest at HighIQ, Nidhi was recently promoted as a team lead and continues to aspire to learning something new every day.