Human Learning and Machine Learning
Learning More Better
Sometimes when I pause to reflect on how much we know about optimizing human learning, I’m just staggered. There’s truly a craft and an art to learning faster, more deeply, and more effectively. Pretty much any endeavor we begin can be enhanced by honing our learning techniques.
To give you a taste, a few resources that come to mind are Cal Newport, Scott Young, Kalid Azad, Dr. Barbara Oakley, Gabriel Wyner, Tim Ferriss, and tools such as Anki or Memrise.
Metalearning has some core similarities across fields, but also some differences. For instance, in my experience, spaced-repetition software like Anki works brilliantly for chemistry concepts or language vocabulary, but not so well for mathematics! For mathematics, a steady deep dive into every single concept that supports the main chunked concept works better.
Dreaming of Electric Sheep
Suppose you’re interested in data science. And why not? Programming lets us do more than ever before, statistics helps us view the world in exciting and more accurate ways, machine learning is revolutionizing field after field.
If you want to help the world, the core skillsets of data science will allow you to do so in many different ways. In the future, for instance, I’d like to apply these skills I’m learning to robotics and biological research.
And the money is good, too!
Well, then, why not pause for a moment to consider the best and fastest way to learn this stuff? What’s the most effective way for a human to learn how machines learn?
Actually that’s easy
Fortunately, several people have already done most of the hard thinking for us.
Jason Brownlee, down in Australia, has a Ph.D. in AI. He’s run a couple of websites for a long time, with incredibly rich and generous content on it. I received his weekly email about biologically-inspired algorithms for awhile.
In this post, Jason explains how to start learning machine learning.
Most people, including myself, might think, “gee, I guess I’d better go learn linear algebra, probability, statistics, calculus, and then I’ll be ready to start.” For the last year, I’ve been taking math courses partly for this reason.
Definitely read Jason’s entire post - in a word, though, he says to just get started. Pick reasonable problems, small problems. Build up from there. Learn the math as you need it.
As you can easily confirm for yourself, if you google “how to learn data science,” most of the responses list huge amounts of resources to go learn. There are books. MOOCs. Blogs. Videos.
Of these MOOCs, the consensus is that Andrew Ng’s Coursera class is the best place to start. It doesn’t require years of math before, and gets you started right off.
But even these MOOCs aren’t the best way to start. Use MOOCs, and practical projects, right out of the gate.
In another post, Jason describes how and why to build a machine learning portfolio. This post backs up the post I described above, because building a portfolio bite by bite is a logical way to learn machine learning. You’ll end up with a portfolio that looks like this, and you’ll know data science.
A structured course
There are also two websites that let you practice - not just learn theoretically, but practice - data science in your browser. The browser-based format means you can get up and running quickly, and it means you can read and practice at the same time.
Both have good reviews, so choose based on your preferred language. Dataquest is Python-focused, and DataCamp is R-focused.
How is this metalearning?
A lot of smarter learning comes down to actually practicing what you need to practice, and avoiding ineffective or misleading study practices.
If you’re learning French, don’t spend all your time review English-to-French translations. Start with a picture of a cat, and practice recalling chat as the translation.
If you’re learning math, don’t spend all your time practicing formulas. Practice deriving and understanding the results and their connections, and the subject opens up.
Data science is a practitioner’s art. The deeper your mathematical understanding, certainly, the more you’ll understand about adapting using the tools correctly, and the faster you’ll understand the firehose of research pouring out right now.
Nevertheless, the aspiring data scientist is much better served by jumping into a small-scale, self-chosen project, as soon as possible. The process of answering a question you want answered gives real-world feedback, and the resulting portfolio displays skills that can’t be faked.
As always, the prequisites are less than you fear.