What is Artificial Intelligence and Machine Learning?
The terms Artificial Intelligence (AI) and Machine Learning (ML) have become popular buzzwords that we hear all the time. But what really is AI? What’s the difference between AI and ML? And most importantly, how do machines truly learn?
Artificial intelligence is the concept of creating machines that can mimic human thinking, decision making and behaviour. On the other hand, machine learning is a subset of AI where machines can learn and make decisions based on large amounts of data rather than being explicitly programmed. There are three types of machine learning: supervised learning, unsupervised learning, and reinforcement learning.
Supervised Learning
In supervised learning, the machine is given a set of input data and the correct corresponding output data. For example, a machine that’s trying to classify if a picture is a dog or a cat will be given thousands of pictures of dogs and cats along with the correct labelling for each picture. Through analyzing the data, the machine can form connections and detect patterns to make accurate predictions on identifying images of cats and dogs that it’s never seen before. This type of learning is the most common and can be seen in a variety of applications including image recognition and spam detection.
Supervised learning is often split into two types: linear regression and logistic regression.
Linear Regression
Linear regression is effective at modelling situations with continuous (real) value outputs. This includes applications such as predicting stock prices and evaluating trends. Let’s use a simple example of predicting the price of a house given its size. We can feed data into the machine on existing house sizes along with their corresponding prices. This data can be plotted as points on a graph and we can find its line of best fit to predict house prices given any house size.
The machine tries to find the line that best fits the data through a process called Gradient Descent. A detailed explanation of Gradient Descent is outside the scope of this article, but essentially, this technique allows the machine to alter its parameters to find the optimal line.
In most situations, there are numerous input variables that all influence an output (e.g. house size, number of rooms and the age of the house all affect the house price). We can account for more input variables in our model through a technique called Multivariate Linear Regression. This allows the machine to take in numerous dependent variables to make a better prediction.
Logistic Regression
Logistic regression is the preferred technique in situations where you have distinct output values such as trying to classify images or predicting whether or not a tumour is cancerous. In logistic regression, the machine tries to predict the probability of something. For example, we can try to predict whether or not a mouse is obese given its weight. In the following graph, several data points display the weight of some mice and whether or not they are obese (where 1 = obese, 0 = not obese).
Rather than fitting a line to the graph, we fit an “S” shaped curve instead, marking the model’s prediction. As you can see, as weight increases, the curve goes up, showing that the machine predicts a higher probability of obesity. The machine works to find the best curve that fits the data.
Unsupervised Learning
In unsupervised learning, the machine is only given input data but no correct output. The lack of a right or wrong answer allows the machine to make its own observations and discover hidden structures, distributions and relationships within the data.
Clustering is a common use of unsupervised learning. The goal is to find objects with similar traits and put them into groups. Say we planted a bunch of flowers. Through looking at the petal lengths and widths, the machine can identify clusters within the data although it was never told which flowers belong to which type.
Unsupervised learning allows the machine to discover patterns in the data that may not be obvious. Applications of this can be seen in marketing. The machine can help group together customers with similar traits allowing a company to send different marketing messages and effectively target each group of customers.
Reinforcement Learning
Reinforcement learning greatly differs from the previous two types of learning. In reinforcement learning, the machine isn’t given any data at the start and gets placed in an environment instead. Through taking actions and receiving feedback, the machine can calculate the value of each action and make optimal decisions.
This form of learning closely resembles how humans learn. After all, we aren’t born with large amounts of data stored in our heads. We learn about the world around us through exploring and interacting with it. The positive and negative reinforcements we receive from our actions give us an understanding of our environment and allow us to make better choices in the future.
The applications of reinforcement learning can be seen primarily in games. In complicated games such as chess or Go, the huge number of possible states of the game makes it nearly impossible to program specific instructions for a machine. In reinforcement learning, the machine gains experience through playing the game hundreds of thousands of times and can use the collected data from those games to learn and improve.
In Conclusion
Machine learning is a powerful tool that impacts our everyday lives. Although this article only scratched the surface, hopefully it has given you a better idea of how machines learn and operate. These developments aren’t stopping anytime soon and I would highly encourage anyone interested to learn more and dive deeper into the cool and crazy world of artificial intelligence.
