Artificial intelligence (AI) is becoming a staple across every industry, with engineers creating algorithms, models, and software to replicate, replace, and supplement human judgment and intervention. Machine learning (ML) falls under the AI umbrella and is on the cutting edge of using data and technology to imitate the way humans learn.
Despite the growing reliance on AI, there is still a cloud of mystery, intrigue, and skepticism about machine learning. At Fiddler, we believe it’s essential to be able to trust the AI solutions ubiquitous in our lives. We also believe that trust comes (at least in part) from better transparency and deeper understanding of AI and ML in general.
To that end, we are demystifying machine learning and model monitoring. In this article, we break down ML and AI basics, including the difference between algorithms and models and the key types of each.
Machine learning is commonly defined as the study of computer algorithms that can improve automatically through the use of data and experience—i.e., learn to perform better. It is important to understand some basics of ML before diving into machine learning model examples.
Essentially, machine learning algorithms build a model based on sample data (also known as training data) to make predictions or educated decisions without being explicitly programmed to do so. A simple example to illustrate this is Spotify. Based on the music you like, listen to, dislike, save, and look up, Spotify curates a playlist of suggestions. In order to do this, Spotify employs machine learning. Without ML, it would be impossible to make data-backed suggestions for every single customer.
As you can probably imagine, there are several benefits to machine learning. Most of them boil down to improved efficiency and automation of complex and data-heavy tasks by reducing the need for human toil.
Machine learning can be broken down into overarching categories, algorithms, and models. Up to this point, we have used those terms relatively interchangeably, but now it’s time to dig into the differences between them.
An algorithm in machine learning is a procedure that is implemented within code and runs on data. A model is the output generated by the algorithm and is made up of model data. In other words, algorithms run a specific type of data analysis or automatic programming, and “model” in this context refers to the program or analysis.
A simple analogy is to think about creating the best fit line (linear regression) in Excel. The algorithm tells Excel what to do, and the graph and coefficients generated are the model that represent what the algorithm accomplished. With both algorithms and models, there are different types of each, so let’s look at those now.
One way to think about ML models is to consider what value they have, or how you can use them. To that end, there are three main types of models. They are:
While there are all sorts of ML algorithms, there are three overarching categories. They are:
In many regards, artificial intelligence and machine learning promise tremendous opportunity and progress for society. But, just like all humans have innate and implicit biases and blind spots, AI is imperfect as well. The problem is that most of what machine learning does happens in a black box, making it difficult to detect model bias or flaws in the process.
At Fiddler, we help your MLOps and Data Science teams develop responsible AI by providing explainable AI. Once you understand why your ML models are making certain decisions, you can improve their overall performance.
Try Fiddler to get started on your path to building trust into AI.