Machine Learning : In this blog post we will discuss about the sub domains and division of algorithms in machine learning. A discussion is conducted in order to put different types and related categories for Machine Learning in perspective.
The evolution of AI has changed the entire twenty first century in terms of technology. A has stolen the spotlight and its advancements are quicker than we predicted. With such an exponential growth in AI, machine learning is becoming the most trending field of the twenty-first century. It is starting to redefine the way we live and it's time we understood what it is and why it matters.
What is Machine Learning?
Machine learning is the science of getting computers to act by feeding them data and letting them learn a few tricks on their own without being explicitly programmed. A lot like a human child, so let's consider a small scenario to understand machine learning. As a child if you had to distinguish between fruits such as cherries, apples and oranges you wouldn't even know where to start, because you're not familiar with how the fruits look. As we grow up we collect more information and start developing the capability to distinguish between various fruits. The only reason why we are able to make this distinction is because we observe our surroundings, we gather more data and we learn from our past experiences. It's because our brain is capable enough to think and make decisions. Since we have been feeding it a lot of data and this is exactly how machine learning works. It involves continuously feeding data to a machine so that it can interpret this data, understand the useful insights, detect patterns and identify key features to solve problems. This is very similar to how our brain works now. Let's move ahead and take a look at the different types of machine learning.
Types of Machine Learning Algorithms
Machine learning can be broadly categorized into three main types based on the learning style and nature of the problem they solve. Each category has its own set of algorithms, techniques, and applications, and they can sometimes be used together or in combination to solve complex problems in machine learning. These categories are listed below:
Supervised Learning
Unsupervised Learning
Reinforcement Learning
First of all we have supervised learning. We all know that supervise means to oversee or direct a certain activity and make sure it's done correctly. In this type of learning the machine learns under guidance. Just like at school, our teachers guided us and taught us. Similarly in supervised learning machines learn by feeding them labeled data and explicitly telling them hey, this is the input and this is exactly how the output must look.The teacher in this case is a training data.
Next we have unsupervised learning. Unsupervised means to act without anyones supervision or without anybody's direction. In this case the data is not labeled, there is no guide and the machine has to figure out the data set given and it has to find hidden patterns in order to make predictions about the output. An example of unsupervised learning is an adult like you and me. We don't need a guide to help us with our daily activities. We can figure things out on our own without any supervision.
Finally we have reinforcement learning. Reinforcement means to establish or encourage a pattern of behavior. Let's say that you were dropped off at an isolated island. What would you do now? Initially you'd panic and you'd be unsure of what to do, where to get food, how to live and so on. But after a while you will have to adapt. You must learn how to live on the island, adapt to the changing climates, and learn what to eat and what not to eat. Basically following the hit and trial concept because you know the surroundings and the only way to learn, experience and then learn from your experience is what reinforcement learning is. It is a learning method wherein an agent which is basically you, stuck on the island interacts with its environment. By producing actions and discovering errors or rewards. Once the agent gets trained it gets ready to predict the new data presented to it.
Other Categories or Subfields in Machine Learning
Apart from the above mentioned broad categories in machine learning there are others which are mentioned below:
Semi-supervised learning is a machine learning paradigm that falls between supervised and unsupervised learning. It leverages both labeled and unlabeled data to improve learning accuracy when labeled data is limited or expensive to obtain. Semi-supervised learning is valuable in scenarios where labeled data is scarce but unlabeled data is abundant, allowing models to make better use of all available information and improve overall performance.
This involves interactively querying the user or an oracle to obtain labels for the most informative data points. Active learning is a machine learning approach that involves an iterative process where the model actively selects the most informative or relevant data points from a pool of unlabeled examples and requests labels for those specific instances. The goal is to reduce the labeling effort by focusing on the most valuable data points for training the model.
Transfer learning is a machine learning technique where knowledge gained from solving one problem (source domain) is transferred and applied to a different but related problem (target domain). It involves using pre-trained models or representations learned from one task to improve learning or performance in another task, typically when labeled data is limited in the target domain.
Deep learning is a subset of machine learning that involves the use of neural networks with multiple layers (deep neural networks) to learn and extract patterns or representations from data. It enables the learning of intricate hierarchical representations from complex and unstructured data, allowing for more powerful and sophisticated learning models.Deep learning has revolutionized various fields due to its ability to handle complex data and learn intricate patterns. Advancements in hardware, algorithms, and architectures continue to expand the capabilities of deep learning, making it a powerful tool in machine learning and artificial intelligence research.
Difference between Supervised, Unsupervised & Reinforcement learning
Let's move ahead and look at the differences between supervised, unsupervised and reinforcement learning. So let's begin by looking at those definitions. Like I mentioned earlier, supervised learning is a type of machine learning wherein we teach the machine using labeled data.Input and output is labeled. Next we have unsupervised learning . The data provided to the machine is not labeled and the machine has to learn without any supervision. That's why it should discover hidden patterns and trends in the data. Finally we have reinforcement learning. The basic concept of reinforcement learning is that there is an agent.This agent is put in an unknown environment. The agent has to explore the environment by taking actions and transitioning from one state to the other so that he can get maximum rewards.
Type of data used to train a Machine
The next parameter is the type of data used to train a machine when it comes to supervised learning, it is quite clear and simple the machine will be provided with a label set of input and output data in the training phase itself so basically you feed the output of your algorithm into the system this means that machine already knows the output of the algorithm before it starts working on it. An example is classifying a dataset into either cats or dogs.If the algorithm is fed an image of a cat. The image is labeled as a cat. Similarly for a dog this is how the model is taught. The model is then tested, using a new dataset but a point to remember here is that in the training phase for a supervised learning algorithm the data is labelld.In unsupervised learning the machine is only given the unlabelled input data. In this case we don't tell the system where to go the system has to understand itself from the input data. Unsupervised learning the machine will be fed images of cats and dogs. At the end it will form two groups one containing cats and the other containing dogs. The only difference here is that it won't add labels to the output. It will just understand how cats look and cluster them into one group and similarly for dogs. Coming to reinforcement learning there is no predefined data. The input depends on the actions taken by the agent. These actions are then recorded in the form of matrices so that it can serve as a memory to the agent. As the agent explores the environment it will collect data which will then be used to get the output. In reinforcement learning there is no pre-defined data set given to the machine. The agent does all the work from scratch the next parameter to consider is training in supervised learning. The training phase is well defined and very explicit. The machine is fed training data where both the input and output are labeled. The only thing the algorithm has to do is map the input to the output. The training data acts like a teacher or a guide. Once the algorithm is trained, it is tested using the new data. When it comes to unsupervised learning, the training phase is vague because the machine is only given the input and it has to figure out the output on its own. There's no supervisor or mentor in reinforcement learning.No pre-defined data and the whole reinforcement learning process itself is a training and testing phase since there is no pre-defined data given to the machine, it has to learn everything on its own and it starts by exploring and collecting data.
Problem solving in Supervised, Unsupervised and Reinforcement Learning
The next parameter to consider is the type of problems that are solved using supervised, unsupervised and reinforcement learning.
Supervised Learning - Problem Solving
The main aim or the end goal of a supervised learning algorithm is to forecast an outcome now obviously that is the basic aim of all these machine learning types but the whole supervised learning process is built in such a way that it can directly give you a predicted outcome. Under supervised learning we have two main categories of problems: we have regression problems and we have classification problems. there is an important difference between classification and regression, basically classification is about predicting a label or a class whereas regression is about predicting a continuous quantity.Let's say that you have to classify your emails into two different groups so here basically we'll be labeling our emails as spam and nonspam mails for this kind of problem where we have to assign our input data into different classes we make use of classification algorithms. On the other hand regression is used to predict a continuous quantity. A continuous variable is a variable that has an infinite number of possibilities. for example a person's weight. someone could weigh one eighty pounds or they could weigh one eighty point ten pounds or one eighty point one one zero pounds. The number of possibilities for weight are limitless and this is exactly what a continuous variable is.Regression is a predictive analysis used to predict continuous variables. Here we don't have to label data into different classes. Instead you have to predict a final outcome. As an example let's say that you want to predict the price of a stock over a period of time. For such problems you can make use of regression algorithms.
Unsupervised Learning - Problem Solving
Coming to unsupervised learning this type of learning can be used to solve association problems and clustering problems. Association problems basically involve discovering patterns and data finding co-occurrences. A classic case of association rule mining is a relationship between bread and jam. People who tend to buy bread also tend to buy jam. It's all about finding associations between items that frequently co occur or items that are similar to each other. Apart from association problems, unsupervised learning also deals with clustering and anomaly detection. Problems clustering is used for cases that involve targeted marketing, wherein you are given a list of customers and some information about them and what you have to do is you have to cluster these customers based on their similarity using a clustering technique to cluster potential buyers into different categories based on their interests and their intent. Anomaly detection on the other hand is used for tracking unusual activities. An example of this is credit card fraud, wherein various unsupervised algorithms are used to detect suspicious activities.
Reinforcement learning Learning - Problem Solving
Reinforcement learning is comparatively different . In reinforcement learning the key difference is that the input itself depends on the actions we take. For example in robotics we might start in a situation where the robot does not know anything about the surroundings. It performs certain actions it finds out more about the world but the world it sees depends on whether it chooses to move right or whether it chooses to move forward or backward. In this case the robot is known as the agent and its surrounding is the environment. For each action, it can receive a reward or it might receive a punishment.
Comments