What is Machine Learning?
Machine learning (ML) is a branch of artificial intelligence based on the idea that machines (anything that has a CPU/GPU or processing system) can learn from data, identify patterns, and perform tasks without or with minimal human intervention. can take decisions.
Why has machine learning evolved so much in the past decade?
There are many reasons for this to happen, although the main reason machine learning has developed so rapidly over the past decade is processing capacity and storage capacity.
If you look in the image above, we can see that the number of transistors on Microchip is doubling every two years. This is related to Moore's law, we will talk about Moore's law in a separate post. However, now we come back to machine learning.
Whereas machine learning algorithms are available from decades ago. However, with the advancements above enable the ability to apply complex mathematical calculations to big data in faster ways than ever before. Here are some examples of machine learning advancements that you may have seen or are familiar with:
Many more industries are implementing Machine Learning or have already implemented it
Difference Between ML and Artificial Intelligence (AI)
AI is a broad science that mimics human behavior, ML is a subset of AI that trains a machine to learn.
Why is ML important?
With advances in technology, we now know that it is possible to quickly produce ML models that can analyze complex data and provide fast yet accurate results at any type of scale.
By building these accurate models, companies now have better chances of identifying profit opportunities and avoiding unknown risks.
What are some popular ML methods?
ML methods are broadly divided into a few categories with supervised learning and unsupervised learning being the most popular. Although there are other categories in ML, let us learn them one by one.
Supervised learning refers to algorithms that are trained using labeled data. For example, at the end of the year a student is given a result as "Pass" or "Fail". In this case, the respective learning algorithm derives the final result as well as the relevant factors for student performance such as attendance etc.
Post which algorithm learns by comparing predicted vs actual output and revises itself accordingly. In supervised learning, methods such as classification, regression, and gradient boosting use patterns to predict the values of the outputs for future classes. It is commonly used in places where we have historical data labeled and we need to predict future events.
Unsupervised learning refers to algorithms that do not contain any labeled data. In short, the "correct answer" is not provided to the machine. The algorithm tries to find out what is shown in the data. Mostly, the goal in unsupervised learning is to trace the data and find out the underlying structure in the data. For example identifying segments of customers with similar characteristics for a marketing campaign.
Think of unsafe in the context of a particular school classroom, if we are not told what we need to find, except try to find similar characteristics. What would you do? Begin grouping students based on their physical characteristics, perhaps by height or weight, eye or hair color, etc. Very quickly you can start identifying characteristics for that particular group.
Another use case of unsupervised learning is dimensionality reduction.
Semi-supervised learning refers to algorithms that use both labeled and unlabeled data. Most follow the 80-20 principle. That is, 80% of unlabeled data and 20% of unlabeled data, mainly because the cost of obtaining labeled data is much higher than the cost of obtaining unlabeled data.
An example might be detecting a person's face on a web cam.
Reinforcement learning - Early use of reinforcement learning began in the fields of robotics, gaming, and navigation. However, it is also developing in other domains. In reinforcement learning, the algorithm learns by trail and error method. This means that the algorithm tries to identify which actions will yield the highest reward. Reinforcement learning has three main components (a) Agent (learners or decision makers), Environment (everything with which agents interact), (c) Actions (what the agent does)
0 Comments