ai
Really, AI is just a term for a program that you can train to perform a specific task more complex than the basic inputs and outputs you see in a lot of coding.
machine learning
Machine learning is a key concept in creating AI. With a machine learning program, a program is assigned a task (for example, group these objects into three groups that are similar to each other, or point out all the pictures that have a cat in them).
algorithm
When the Youtube algorithm suggests a new video, it is looking at all the data it has on your search and watch history, and trying to choose videos that will get you to keep watching as long as possible. The problem the algorithm is solving is maximizing your watch time, and it solves this problem through a series of complex programming equations.
neural networks
Let’s say you want to buy a book to buy for your sister. Many factors impact that decision: how old your sister is, what books she has read before, what authors she likes, etc. You weigh all of the factors in this decision differently, and make a choice. Neural networks can work similarly. You give it data with a number of variables, just like the ones in the book example, and a series of algorithms work together to make a decision that they think is most accurate or successful.
imputation
Sometimes the data you’re working with has pieces missing. If you have lots of data in your data set, you may be able to delete any data with missing pieces and still get pretty good insights. But if there is a lot of missing data, or if you can’t afford to throw some data out because there isn’t a lot to work with, you can use data science programs to make an educated guess about what the missing data would most likely be.
supervised learning
In supervised learning, I can use “training data” to teach my model what I’m trying to predict. For example, if I run a beauty brand and I wanted to predict which customers are most likely to buy my new products, I could give the model information on all of my previous beauty products, and some information on the customers that bought them - age, location, how often they visit my website, and any other data I’ve gathered. The model will then use that data to understand which customers buy which products most. Then, I can give it more data (“test” data) to practice making predictions on, and use that to understand how accurate the model is.
unsupervised learning
In supervised learning, I’m looking to get a specific answer from my model but in unsupervised learning, I just want my model to explore the data looking for insights. For example, if I gave an unsupervised learning model information about all the people who attended Comecon, it could group the attendees that seem most similar to each other. With this, I couldn’t predict anything, but I could understand that the event attracts teenage boys who like Sailor Moon or women from Boston.
nlp
Not long ago, Siri couldn’t understand anyone speaking English with a non-American accent. Natural language processing is the field of machine learning making machines better at things like this.