A Roadmap to Deep Learning?

Topic of the Week

Note: I would appreciate any comments or insights on this topic. Use the comment section at the bottom or respond to the tweet here.

I didn't particularly want to talk about deep learning again as I've built a list of topics for reflection. However, after some thought, I found that the title of this blog post, "A Roadmap to Deep Learning", was a catchall for most of the topics.

Potential Blog Ideas (Pillay 2022)

The idea for this week's blog came from a mixture of sources. I hope that the overall post shows sufficient reflection, despite mostly being a sort of annotated bibliography.

First was the idea of deep learning, which has been on my mind since we covered it in a lecture. I’ve experienced an accompanying feeling of overwhelm at the idea of deep learning, how to do it beyond “become absorbed in the content”. Our twitter assignment has revealed many, many useful resources which I have collected and organised but it has been impossible to become deeply absorbed in much of it, if any.

While multitasking in order to get more quality work done in a given time is kind of a myth, I find the feeling of tasks piling up when focused on a single task unpleasant. I particularly dislike when I don’t have a clear idea of a task’s scope and the steps required to accomplish it, which caused a fair bit of panic when working on my proposal for my summer project.

Multitasking (James Clear 2022)

The second source was a talk by Brandon Sanderson, a famously prolific fantasy writer (he recently announced that he had written 5 new "surprise" books during the pandemic, on top of his usual work of writing his fantasy epics, the latest of which was 455,000 words). The talk, titled The Common Lie Writers Tell You, was the keynote address at the 2020 YallWest Book Festival, and was about how to achieve hard things. Subtopics included his teaching philosophy and the idea of creating a detailed roadmap for achieving goals.

Third, was the YouTube channel The OverAnalyzers. It's a small channel that I found via one of the host's old music theory videos. The hosts are brothers, and they discuss evidence-based or popular methods to improve themselves in some way. What I like about the podcast is that they put the concepts into daily practice, usually for a month, and then discuss what they learned "on the ground". It isn't quite as evidence-based as, for example, Andrew Huberman's channel, but their experiments, goals and challenges give an interesting insight into the topics.

Fourth, was the revelation of doing "extra" courses to learn various skills that aren't covered in our masters programme. I have been doing a free two-week course with FutureLearn this week titled How to Create an Online Course and it has been very beneficial in revising, consolidating and adding to some topics I've encountered during this Masters. This also applies to various YouTube videos, which provide opportunities for similar learning, albeit in a more passive format.

How to Create an Online Course (FutureLearn 2021)

Fifth is the idea of mental models, which are linked to schema theory, a practical way to view a topic or problem, aided by what we already know.

Sixth, was the feeling that, when designing a course, there are so many things to think about. Language, design, inclusivity accessibility are all large topics that can be broken down to minute detail and would benefit from mental models when applying them. Checklists may be a good format for these mental models.

Last, was a combination of my use of a simple checklist app to keep track of what I needed to achieve this week and Atul Gawande's book, The Checklist Manifesto: How to Get Things Right, which discusses of the impact of checklists on the medical, construction, and aviation industries.

The Checklist Manifesto (Atul Gawande 2009)

Putting all that together, I've decided to attempt to create a personalised Roadmap to Deep Learning.

My first thought was that I believe the roadmap will need to be specific to the topic (i.e., this masters, or a section of it such as web/graphic design), but I will probably learn a lot along the way about how I can generalise or simplify it. I also will concentrate on making it for myself; the goal of creating a roadmap that anyone can use, for any topic, is way too big.

Although I’ve learned about many learning theories, design fundamentals, and many other topics during this masters, I don’t feel that I have a grasp of how to use the information. The topics and theories are separate entities and I would like to form them into a usable whole, and I feel that my fascination with the concept of deep learning is significant.

Currently, I see it as being a linear (or lightly branching) process diagram with each node containing one or more checklists. I aim to create one checklist at a time as I encounter various topics or skills that I need to learn. The checklists should only contain declarative statements of actionable steps.

This possibly all sounds silly. I think it could be a fool's errand and I know that this may reinvent the wheel to an extent, though a brief google search revealed disparate tools for deep learning, but not a path. I currently know nothing about process development in companies and industries which would probably be helpful. This may just end up being a curated collection of checklists from various sources and I’m uncertain if that would actually be “a roadmap to deep learning”. I will at least learn a lot along the way.

I’ve come up with some rules, so this doesn’t interfere with my studies and other goals/responsibilities:

1. This project cannot delay my assignments.

2. This project cannot delay my other goals, which are documented at the end of each post.

3. Create simple checklists for current assignments, more detailed checklists after reflecting on the assignment and how deeply I learned.

4. No experimenting with tools for creating the roadmap for the moment - create checklists only.

All comments and suggestions are welcome.

I will share my progress in future posts.

Until next week.

Last Week's (and This Week's) Next Steps

• Research jobs, including required competencies and ideal portfolio content.

• Recreate/update CV.

• Continue with the goal of learning deeply.

• Continue lifestyle changes.

• Study as many online courses and e-portfolios as possible and keep record/screenshots of good ideas.

References

Brandon Sanderson (2020) 'The Common Lie Writers Tell You' YallStayHome 2020 Afternoon Keynote [video], available: https://www.youtube.com/watch?v=oH9sJrAVeC0 [accessed 4 Mar 2022].

FutureLearn (2021) How to Create an Online Course - ELearning Training Course, available: https://www.futurelearn.com/courses/how-to-create-an-online-course [accessed 4 Mar 2022].

Gawande, A. (2009) The Checklist Manifesto: How to Get Things Right, New York: Metropolitan Books.

Le Cunff, A.-L. (2019) ‘30 mental models to add to your thinking toolbox’, Ness Labs, available: https://nesslabs.com/mental-models [accessed 4 Mar 2022].

Clear, J. (2022) 'The Myth of Multitasking: Why Fewer Priorities Leads to Better Work', James Clear, available: https://jamesclear.com/multitasking-myth [accessed 4 Mar 2022].

Clear, J. (n.d.) ‘Mental Models: Learn How to Think Better and Gain a Mental Edge’, James Clear, available: https://jamesclear.com/mental-models [accessed 4 Mar 2022].