September 17

0 comments

How to Avoid Getting Lost Down the Machine Learning Rabbit Hole (The T-Frame Concept)

In 1862 Charles Dodgson and Reverend Robinson Duckworth rowed up the Isis river with 3 sisters. While on the trip Charles told the girls a story about a bored little girl who went looking for adventure and they loved it. He continued to elaborate the story and two years later handed one of the girls the completed manuscript in featuring her name. The following year Charles published Alice's Adventures in Wonderland using the pseudonym of Lewis Carroll as author.

We probably all know the story about how Alice enters Wonderland very well. She falls down a rabbit hole doesn't she. And so began her adventure into a logic defying underground world of nonsense and fantasy. Yet do you remember how Alice found her way back from her adventures into Wonderland? After surviving many challenges and crises she found her way back out of the rabbit hole.

Starting out in the field of machine learning can feel very similar to falling down Alice's rabbit hole. The world of machine learning is very deep and there are many different tangents to explore. When starting out it is natural that such a deep topic can feel a little daunting. So much so that you might feel like you struggle to make progress for quite a long time. You can easily spend months just skimming over the high level topics without actually building a machine learning model. And just like Alice, the challenge is even greater if you are entering the machine learning world on your own. In this article we will discuss how to find your way through the mind stretching world of machine learning without getting lost down the rabbit hole.

This article will cover the topic of how to design your own machine learning curriculum and is divided into three parts.

  1. Outlining your journey
  2. Pick one
  3. The T frame

At first you outline the topics you are interested in

One simple way to begin is to open a classical textbook such as Artificial Intelligence: A Modern Approach by Russell and Norvig. If you follow the link you'll find the contents page of this book which is used in computer science departments around the world. You can quickly get an outline of the major topics in AI and where machine learning fits into this broader topic without needing to even purchase the book (although if you are serious about machine learning it's probably a good idea). A contents page can provide a useful outline of the important topics available for study. And it also provides a map to return to if you find yourself getting lost down one of the rabbit holes.

And this quick outlining technique can also be applied to websites as well. For example, influential AI researcher, Eliezer Yudkowsky in his blog, Less Wrong, has a lovely summary of AI concepts. One pro tip you can leverage when you find such a treasure trove of interesting topics is to save the page with links and all to Evernote. This can then be used as a dashboard to jump into any number of rabbit holes and know that you can still get back to the place you began. If you find yourself lost down a rabbit hold just return to the dashboard. So outlining can be done very quickly but that's still quite a lot to digest which brings us to the next part, where we make a sense of these topics by picking one.

Where should you start in machine learning?

The answer "it doesn't really matter" is probably true but not very instructive. Instead you are best advised to pick the topic that interests you most and dive into that. This might mean diving into a topic such as reinforcement learning on your first day. Even though this is a very advanced machine learning topic don't let that stop you from learning as much as you can about the subject. You'll quickly discover the things that you need to know and when things start getting too much. If you start getting lost it's not your fault.

It would be a mistake at this point to let the feelings of intimidation deter your learning journey. Maybe you started reading about reinforcement learning because it sounds most exciting. Perhaps you first became interested in machine learning after hearing about how AlphaGo used a deep reinforcement learning algorithm to defeat the world Go champion, Lee Sodol in 2016. It's very important to encourage your curiosity at this stage. The topics that interest you most will provide fuel for your journey. It's important to realise that when you start to feel a little lost it is a reminder that there are possibly some basic concepts you haven't learnt about yet. And this brings us to part three in your self guided journey into machine learning and the T frame.

Browse widely on topics then go deep into one (The T-Frame)

We are calling it the T frame because it helps you keep your learning in something that can more easily be handled like a picture frame. You can imagine the top of a capital T as the high level topics you discovered in the outlining you completed in part one. When you picked one topic like reinforcement learning you started to go down the vertical on the T. But very quickly your hackles may be raised. Concepts that are unfamiliar can become intimidating. When you find this sense of intimidation creeping in it's not a sign to run away. Quite the opposite.

This is when you prepare to go deeply into a topic. But there are some topics you may not be ready for because everything is just too much of a stretch. That's when you remember your T frame. You may not be ready to build your own version of AlphaGo, at least not yet. Instead, you can return to the top level of the T and browse back to an earlier topic such as linear regression that feels more accessible to you. You can still have lot's of fun with linear regression and machine learning. You haven't given up on your dreams and inspiration at all. Not at all. Your early shallow explorations will provide you with ongoing inspiration to learn more. And the idea of the T frame allows you to return to those rabbit holes again in the future when you have armed yourself with a deeper understanding of the necessary concepts.

And already we've covered the 3 parts. Let's do a quick summary.

  • First find a popular machine learning or AI reference book and survey the contents pages.
  • Store these contents and topic summaries from websites in Evernote with links.
  • The outline is your learning dashboard. It's the horizontal line on your T frame.
  • Sample the topics that interest you most and discover the concepts that are unfamiliar.
  • Pick one concept and dive deeply into learning that one thing. This is the the vertical line in the T frame.
  • The T frame allows you to find your way back out of the rabbit hole if you ever happen to get lost.

Next Steps

This is the first in my series of how to get started in your learning journey into artificial intelligence and machine learning.  In the next article we will focus on how to create a machine learning study plan for yourself that is actually fun to use.

P.S. Did you remember how Alice got out of the rabbit hole? She was woken up by her sister!


Tags

ai, machine learning


You may also like

{"email":"Email address invalid","url":"Website address invalid","required":"Required field missing"}
__CONFIG_colors_palette__{"active_palette":0,"config":{"colors":{"98315":{"name":"Main Accent","parent":-1}},"gradients":[]},"palettes":[{"name":"Default","value":{"colors":{"98315":{"val":"rgb(0, 165, 179)"}},"gradients":[]},"original":{"colors":{"98315":{"val":"rgb(47, 138, 229)","hsl":{"h":210,"s":0.77,"l":0.54,"a":1}}},"gradients":[]}}]}__CONFIG_colors_palette__
__CONFIG_colors_palette__{"active_palette":0,"config":{"colors":{"9a854":{"name":"Main Accent","parent":-1},"aed69":{"name":"Accent Light Transparent","parent":"9a854","lock":{"saturation":0,"lightness":0}},"0b7bf":{"name":"Accent Dark","parent":"9a854","lock":{"saturation":1,"lightness":1}}},"gradients":[]},"palettes":[{"name":"Default","value":{"colors":{"9a854":{"val":"var(--tcb-local-color-98315)","hsl":{"h":210,"s":0.78,"l":0.01,"a":1}},"aed69":{"val":"rgba(255, 255, 255, 0.18)","hsl_parent_dependency":{"h":210,"l":1.22,"s":1.01}},"0b7bf":{"val":"rgb(34, 46, 62)","hsl_parent_dependency":{"h":216,"l":0.19,"s":0.29}}},"gradients":[]},"original":{"colors":{"9a854":{"val":"rgb(19, 114, 211)","hsl":{"h":210,"s":0.83,"l":0.45,"a":1}},"aed69":{"val":"rgba(60, 152, 247, 0.18)","hsl_parent_dependency":{"h":210,"s":0.92,"l":0.6,"a":0.18}},"0b7bf":{"val":"rgb(35, 47, 65)","hsl_parent_dependency":{"h":216,"s":0.29,"l":0.19,"a":1}}},"gradients":[]}}]}__CONFIG_colors_palette__
Sign Up

Subscribe here to receive updates about machine learning and new posts in your email.  You can also email me directly to ask for help and advice.  

>