In the dawn of the digital age, data is the modern currency. Businesses rely on consumer data to enhance their growth. In the age of automation, the system collects data and, processed, and then used as a tool for data analysis. Whenever you visit a website and spend some time over it. Web-Cookies collects data describing your browsing behavior in a particular application. Google knows your location all the time. E-commerce platform keeps track of the choices that you make or even those that you didn't make. It helps them to offer user attractive and enticing deals on the products that you desired to purchase. Day to day the science of data is advancing, making machines more capable and independent.
Machine Learning is the science of enabling machines to learn on their own without being emphatically programmed for each distinctively similar tasks.
Lets further break down the word Machine Learning. The machine would be any programmed computer or SOC (System on Chip). Learning whereas indicates the degree of smartness a particular system achieves based on the gained knowledge. For any hardware, its degree of smartness lies in how well i'ts programmed.
Say a system needs to determine whether the input picture provided is of the boat or not? Now there are two approaches to observe this problem.
Learning is an intelligent byproduct developed on the basis of past experinces. This is how we as humans learn and grow. This is the approach an efficient machine learning model suffice.There are three key points to keep in mind while understanding an ML approach i.e. Experience, Task, Performance in the particular order.
Therefore, we can define machine learning more formally,
"A model is said to be an efficient Machine Learning Model if and only if a computer, on the experience based on the past task improves it's current performance while conducting a new task,on it's own."
After looking up to a problem ask yourself three important questions-
Learning is successful when a consistent improvement reflected after each trial. For prediction and recognition, we frequently use Machine Learning algorithms. At first, know the purpose of predicting a solution for a particular task? Further, based on the provided data, lay down all the possible approaches that can be applied to fulfill the aim. Finally, identify the most suitable method for a working ML model. The higher efficiency of our machine learning model increases the probability of predictions.
Learning algorithms are working almost everywhere in this modern data-dependent world. Whether it is Database solutions (Medical records, human disease detection) or Product recommendation systems (Amazon, Netflix), machine learning is everywhere.
Some real-life examples of machine learning are:
Further, it is very much important to understand the demands of a particular learning problem. Next, we will discuss the types of machine learning.
Developing a machine learning model requires understanding the nature of the problem. There are numerous ways to look up to a task but only one efficient way to resolve it. The performance of an ML model lies upon the efficiency percentage i.e. possibility of success on the test data. The performance of the efficient learning model improves with each iteration.
The important question is What should be the process of learning? The difference is in the approach to make a model that enables a machine to learn itself. Let's envision it with a relative example.
On this note we categorize machine learning. There are mainly three types of machine learning - Supervised Learning, Unsupervised Learning, Semisupervised Learning and Reinforcement Learning. I will breifly define them here and while proceeding later in the course we will understand them in much detail.
In Supervised Learning, we know the possible outcome. We adequately provide historical data to the standard algorithm and get the desired result. We trained the model on a labeled dataset. The labeled dataset means we have both input and output parameters.
Image Classification represents a classic example of Supervised Learning. The algorithm trained enough to identify the image accurately. When the user provides the image to the system, based on its data, it will recognize the type of image.
Supervised Learning Algorithms
Neural Networks (most of them)
In unsupervised learning, the model trained using unlabeled data. The trained algorithm model acts on the learned information it has and proceeds without explicit guidance. Unsupervised learning algorithms intentionally allow the model to perform complex tasks compared to supervised learning. It also allows the system to self-improve and provides more accurate results over the period.
Let's understand it with the help of an example. Let's say you have a car. And your kid recognizes it. One day you decided to go on a long drive. During the journey, the child yet identifies the vehicles driving on the way without your help based on the information he accumulates from home (like four tires, steering, wipers, etc.).
It is unsupervised learning.
Unsupervised Learning Algorithms
Supervised learning is often time-consuming because it expects to label the data every time. In practical scenarios, it is not possible to always have a labeled dataset. We have partially labeled data. When we train the model using the partially labeled data points, it is traditionally called semi-supervised learning.
Voice Recognition is a classic example of semi-supervised learning models. Labeling audio files remain a time-consuming and highly resource-intensive task. Using the semi-supervised learning improves the accuracy of the voice recognition model.
In Reinforcement Learning system learns from experience. For every action, the system gets the rewards or penalties based on the move. If it has an adverse effect, then the system avoids it next time. All advanced robots implement Reinforcement Learning Algorithms. There is no concept of label or unlabeled data but an initial state from where the model begins.
These are very brief ideas about the types of machine learning. The deeper insights shall be explained in subsequent sections.
In a typical Machine Learning Project, we need data and a model to train. Eventually, we applied the model to predict, whenever a new case or data is feed into the system.