What’s the Difference between Data Science and Deep Learning?

Deep learning is a type a data science.

With data science, you find the probability that something is true using programming techniques. This is a broad definition. Some examples are categorizing pictures of cats and dogs: in this case, you find the probability that a picture is 99% cat and 1% dog. It can be used to find patterns like “songs that this person will like” or “places where people will need a ride.”

There are many ways to arrive at this kind of truth. Deep learning is a very specific technique for finding that probability of truth. It works basically by training an algorithm on a known data set to find probability of truth accurately for unseen data. This training works by calculating a “loss function” — that is, how wrong a random guess is compared to the truth of the known data set. Then, the algorithm — which consists of randomly weighted variables, has a variable slightly increase or decrease in some direction. Based on this change, you can actually see if the loss function is getting bigger or smaller. Using that information, you change the variable in the direction of making the loss function smaller. You repeat this (something simple enough that it can be done in Excel) a few million times and you can arrive at the truth.

To see if you’re interested in deep learning and you’re new to it, check out our podcast, Startup Data Science.