In Layman's Terms
In Layman's Terms
You might not realize it, but machine learning is all around us. For example, Google Maps uses machine learning (ML) algorithms to recommend the best route, Netflix uses it to recommend shows you might like, and Spotify leverages ML to curate playlists just for you. You may have also talked to an intelligent chatbot powered by machine learning. While machine learning solutions seem simple to end-users, they are typically very complex when you look behind the surface. So what exactly is machine learning? How does it work? And why is it important? Let's get into it.
Machine learning is a branch of artificial intelligence (AI) that enables applications to become more intelligent over time without being explicitly programmed to do so. So, while a basic chatbot has to be told how to respond to a specific input (for example, a keyword), ML algorithms learn through experience. Machine learning algorithms are fed extensive amounts of historical data to become more accurate at predicting outcomes.
The training data we use for machine learning algorithms significantly impacts accuracy. For example, a study by MIT found that facial analysis technologies had higher error rates for minorities, especially minority women. Why? Because these groups were unrepresented in the training data (the ML model had less experience analyzing these faces).
Data scientists feed labeled training data into the algorithm and set the perimeters for a pass or fail. The model will then make predictions on new data, and the data scientists will correct any wrong predictions until the model reaches the desired level of accuracy. Image classification software is an excellent example of supervised learning. For example, if you wanted the software to classify whether an image contained a dog or a cat, you would label pictures of dogs and cats in the training data.
With supervised learning, we always start with a specific goal (accurately labeling a dog or cat), but with unsupervised learning, we might not know exactly what we're looking for. Instead, unsupervised models look for patterns in the data. A real-world example of unsupervised learning is finding customer segments. So, an algorithm might analyze many thousands of data points to find connections between people. Like, for instance, that men between the ages of 19 and 23 living in urban areas are more likely to buy action games. Customer segments are valuable to companies because it means they can successfully target specific groups with advertisements.
As the name would suggest, semi-supervised learning is a mix of both supervised and unsupervised learning. The algorithm will make decisions independently and look for patterns, but the data scientists will correct the algorithm if it goes too far off course.
In reinforcement learning, data scientists guide the algorithm through a multi-step decision process by giving it positive or negative cues to help it make the correct decision. The algorithm still decides which path to take, but it's helped along the way.
We generate large amounts of data every day, and this data holds many valuable insights that could improve our daily lives. We couldn't possibly analyze all this data manually, and that's where machine learning algorithms come in. Machine learning is the key to unlocking the next generation of powerful applications, whether in detecting disease or financial fraud, optimizing manufacturing, or offering more relevant suggestions in retail or entertainment.
For more layman's guide computers and programming visit the Coding section of our website.