Learning machine learning is kind of dry

San Francisco, we have a problem.

laptop screen with code
Photo by Caspar Rubin on Unsplash

I love the idea of machine learning. Concepts like cost functions and gradient descent are relatively easy to grasp, and gets only a bit trickier during implementation (especially with all that matrix operations in Octave instead of the loops I’m so used to using as a software developer).

The overarching idea of getting to the best possible model by training it on a growing data set is intuitive and fascinating. We found a way for machines to learn!

But the machine learning course I’m taking, the one that in 2017 is the Gold Standard for an introduction to the field (Ng’s Coursera course) is starting to feel dry to me at week 3. There is still 9 weeks to go.

I’m disappointed by my response to the course. I really wanted it to be something I could feel excited about! But working through the material this week felt laborious. And like any millennial I searched the web for people who might be or have felt the same way I do. I ended up discovering only one or two people who are in the same boat. Lost an opportunity to be motivated by a stranger there.

Well, my search didn’t exactly meet a dead end. While scouring the internet for motivation to carry on the course, I stumbled upon a great resource called The AI Podcast. Prof Andrew Ng who is the instructor for the course I’m taking was interviewed on Episode 32. The short 30-minute interview offered a glimpse into the prospects of being an AI practitioner (developer?) in the near future. As Prof Andrew spoke, I felt my curiosity in the field being revived again. But how long this renewed interest will last is anybody’s guess.

Conceptually fascinating but technically dull

What is one supposed to do when he finds machine learning conceptually fascinating but technically dull?

My goal is to convince myself to finish or drop the machine learning course now. I’m going to list a few pros and cons for myself below.

Reasons to press on with the course:
A young field. When one of the pioneers of deep learning says there’s a lot more work to be done, you know there are massive opportunities in the horizon
Many potential applications. Computing power only recently advanced to the point where individuals can run ML algorithms at home, which means we are likely at the bottom of the hockey stick curve of tinkering and novel applications
Good career prospects. I agree with Prof Andrew’s prognosis that AI is going to be the new electricity, and practically all industries will need to implement aspects of it for business as usual in the future

Reasons to drop it and try a different field:
Implementing ML is quite dry. Data scrubbing, determining optimal learning rate and testing different modelling equations is not exactly super interesting
Opportunity cost. What if (yes, that insidious term) I could spend my time better by getting good at a speciailisation within software engineering that has more apparent appeal to me? After all, we do tend to like what we’re good at

These are incomplete thoughts at this point. For now I’ve decided to press on, even if it is fuelled by somewhat grandiose ideas about the future. In recent years I’ve learned to never let a good dose of inspiration go to waste – always favour action!

The AI Podcast Ep 32, an interview with Professor Andrew Ng

Here’s the specific podcast episode I was talking about!