Fundamentals of Machine Learning for Beginners

Machine Learning (ML) is a crucial subset of Artificial Intelligence (AI) that is present in nearly all digital technology. It is built on a series of layers that contain weighted data to help train models on accurate decision-making, challenging machines to learn from experience rather than relying on explicit programming. 

In today’s modern world, examples of machine learning can be found nearly everywhere from video recommendations on Youtube or Netflix to spam email filtering and predictive texts. It’s nearly impossible to ignore the importance of ML in our daily lives, but what is machine learning?

The Concept of Machine Learning

The history of machine learning first began when Alan Turing suggested the concept of an “ultimate machine” which could compute results based on inputs, but it wasn’t until the Dartmouth Conference in 1956 that the term Artificial Intelligence was first coined by John McCarthy. However, science fiction had already been telling stories of scientists creating sentience for over a century with novels like Frankenstein becoming world renown. 

It wasn’t until the mid-1980s that significant progress started to be made. Although rule-based models known as expert systems were making significant contributions to AI, technology was too limited between the 1960s and 1970s to explore this further, which lead to a shift in focus towards data-driven models, the basics of machine learning. 

Types of Machine Learning

By being fed data, ML models can spot patterns to assist in their decision-making instead of relying on traditional programming that puts a strain on computational resources from an over-abundance of rules. Instead, programmers introduce different datasets to help machines come to their own conclusions. There are three ways this can be done:

1. Supervised Learning: Machine learning algorithms are given a dataset in which they know all of the correct outputs (known as Labels). Similar to how a child might use training wheels when learning to ride a bike, the ML model maps input data to their corresponding outputs until the machine becomes proficient at the tasks.

2. Unsupervised Learning: ML models are given datasets without labels and are required to find patterns within the data in order to find the right outputs. The method requires ML algorithms to develop reasoning which can be implied from datasets curated by programmers who determine the complexity of the data. 

3. Reinforcement Learning: Algorithms expand understanding through repetitions in an environment that rewards or penalizes the machine based on the accuracy of its results. Over time, the algorithm begins to develop its own strategies. 

Machine Learning Algorithms

Algorithms are essential to machine learning. Within an AI model is a collection of different algorithms that are used for processing that all belong to one parent algorithm (like Transformer in GPT’s case). These algorithms use a variety of techniques to analyze input information for better output results including: 

1. Linear Regression: A straight, linear line that is created by a machine as it passes through a dataset graph. Only information for the x-axis (called the Independent Variable) is given, leaving the machine to predict the y-axis (Dependent Variable).

   The machine calculates where best to draw this line through a process called Least Squares Regression. Next, the machine measures data from the x-axis to the regression line, marking its placement on the y-axis. 

   Examples of linear regression can be found in financial software used to forecast price fluctuations in real estate markets and also helps predict crop yields for farmers looking to maximize harvests. 

2. Logistic Regression: A binary algorithm that formulates predictions based on data that is either true or false. Similar to linear regression, the machine draws a line through the dataset that can be used to measure Y. However, instead of a straight line, the line is an S as the binary inputs are only 1 or 0. 

   Logistic regressions are most commonly seen in situations where information is either a yes or no such as in healthcare where symptoms are normally checked off upon discovery or in marketing where teams filter target audiences. 

3. Decision Trees: A branching list of options like a dialogue tree in a video game created by a machine, starting with the root node. There are two types of trees that an algorithm can create: a classification tree or a regression tree.

   Regression trees can be simplified as if-then statements that attempt to make a prediction. For example, “if the man runs 2 miles every day then he will lose x amount of weight or y amount of weight.” Here, the regression tree offers 2 likely predictions based on the original statement.  

   Category trees can use both numerical (continuous) and categorical (discrete) data to classify information. The tree begins with only portions of the given variables to begin making basic connections before adding additional context to cover more complex mapping. 

Machine Learning Tools and Languages

When building ML models, programmers tend to rely on common languages like Python and R. Python is popular because of its simplicity and large community that has developed rich libraries of code including NumPy, Pandas, Scikit-learn, TensorFlow, and PyTorch. 

R is also a popular programming language for developing ML models because of its use in statistics and data analysis. Many of the algorithms explained earlier share their foundations in mathematics, making R a preferred choice for developers.