Learn Machine Learn: Enter the Era of Intelligence!
A comprehensive overview of machine learning and its applications
Hello there. I have been caught up with my junior year of high school and got delayed posting this blog. Usually, my blogs are physics and math oriented. However, my other passion is computer science/programming and I am especially intrigued by Artificial Intelligence (AI) and Machine Learning (ML). In this blog, I will explain these concepts that I covered in my school club along with their applications in the real world and the field of physics.
AI and ML are very much prevalent in our everyday lives. Common uses of machine learning are internet search engines, email filters to sort out spam, personalized recommendations on websites, fraud detection in banks, and lots of apps on our phones such as voice and image recognition. Nowadays, machine learning is applied to much more advanced situations like self-driving cars, sales, and finances for big business, and also for the advancement of science. In fact, one of the cooler uses of ML is auto-imaging. A program called DALL·E 2 can do this as seen below.
A commonly asked question is what is the difference between AI and Machine Learning? In the sections below, I will try to demystify these.
What is Artificial Intelligence?
Artificial Intelligence is quite literally intelligence exhibited by machines. AI is an umbrella term used to describe computers exhibiting a simulation of human intelligence. AI can be as simple as one line of code that allows a user to play pong against a computer. AI programs have been around for a while. In fact, the earliest successful AI program was written in 1951 by Christopher Strachey. So what drastically changed between then and now? Earlier AI systems were only as smart as the inputs and the rules that the user fed to the system. It could not learn or find new ways to improve its results.
computerPaddle.y = ball.yWhat is Machine Learning?
Machine Learning is a subfield of Artificial Intelligence. Machine Learning aims to teach computers certain tasks with data, without explicit programming. For example, if we feed a computer data about a chess game like the moves made during the game and the outcome, then we can train the computer how to play chess and win games. This data along with an ML algorithm will help the computer learn how to play chess and win all future games against any player.
Machine Learning Methods
But how exactly does a computer learn? The obvious answer is a lot of math but first, there must be a conceptual understanding in order to start implementing the program. There are three different learning methods that ML uses which are supervised learning, unsupervised learning, and reinforcement learning.
Supervised learning is teaching a computer the desired behavior with labeled data. For example, an email filter to sort out between inbox and spam would be supervised ML. This is because we have given the computer-specific labels to sort between. Unsupervised learning is teaching a computer to sort between data without any specific labels. In unsupervised ML, a computer has to make its own labels and categorize data into these labels. Lastly, Reinforcement learning is when you teach the computer a behavior based on rewarding desired behaviors and/or punishing undesired ones.
Machine Learning Techniques
Now let’s talk about some of the common Machine Learning techniques used for these types of learning methods. These include Classification, Regression, Clustering, Anomaly Detection, Dimensionality Reduction, etc. to name a few.
Supervised: Classification and Regression
Some common Supervised ML techniques are classification and regression. Classification is when a computer sorts data and classifies it into certain categories. The email filter is a good example of this where emails are classified into spam and not spam.
Regression is a technique used to determine the relationship in the data. For example, you can determine the relationship between the temperature outside and the sale of ice cream using regression. When we visualize this, we typically see it as a best-fit line in a scatter plot. Regression is often used in the field of Astrophysics. In the graph below we can see that there is a positive correlation between the stellar mass of a galaxy and the mass of a supermassive black hole. This relationship (the best fit line) can be determined through an ML regression algorithm.
Unsupervised: Anomaly Detection and Clustering
Some common supervised ML techniques are anomaly detection and clustering. Anomaly detection is when a computer analyzes data and is able to determine whether or not there is data that does match with the other. This technique is very important in the field of physics where some of the most important theories were proven true through ML analogy detection algorithms. One example of this is the discovery of the Higgs Boson which was detected by finding an anomaly in the data. In the graph below, there is a peak between 120 GeV and 130 GeV. Using an anomaly detection algorithm, the ATLAS project was able to find this peak and conclude that the Higgs Boson was 126 GeV.
Lastly, clustering is when the computer groups data that is unlabeled. Clustering is the partitioning or grouping of similar data points. This is supposed to simulate the natural tendency of humans to group objects. Some examples of clustering in the real world are based on sales history, buying patterns, and fund behavior. There are many types of clustering like partitioning, hierarchical density-based, grid-based, and spectral.
Final Thoughts
This blog provides a conceptual understanding of AI and ML, covering the foundations and basic understanding that everyone should be aware of, as we go into this era of intelligence. In my next blog, I will dive deeper into hands-on topics on how these techniques work. We will cover examples of how computers cluster data with methods like K-Means along with how simple it is to code clustering in Python.
