What is Doing Deep Learning “Like” — Is It Similar to Coding or Different?

You’ll be working in a Python Jupyter notebook to do deep learning. So as you’re doing deep learning, you will be coding the entire time. Just like with coding, you’ll want to think carefully about what you’re trying to do before you even try to code it. Similarly, there are existing libraries that you can use to make your job a whole lot easier.

What is a little special is that you will be coding specifically according to the “tradition” of deep learning. Because of how deep learning works, your code should be structured in a certain way, just as web frameworks encourage coders to structure their code according to MVP — it’s not absolutely necessary, but it makes things a lot simpler to work with.

Also, with coding, you tend to use a lot of domain knowledge to write good code. With deep learning, you really only need two types of knowledge — how to structure your data (very simple) and how to experiment with different deep learning techniques. In other words, you don’t need to know a lot of domain knowledge outside of standard deep learning techniques to achieve world class results. For example, Jeremy Howard used deep learning to achieve world class results in identifying cancer despite never studying Medicine unlike teams of PH.D Science & Medicine specialists who studied Cancer for decades whose results Jeremy outperformed in a few months.

In actual practice, deep learning takes a lot of time. So you’ll learn to build your model on a small sample size, but there are still some things things only work with larger data sets. These longer feedback loops mean you want to think a little extra and experiment more with smaller data sets before deciding to run something and wait for results.

Learn how to do all this and more by listening to Startup Data Science.