I’m a big advocate for project-based learning. I share why in this week’s reflection below.
To have impact with machine learning models, we need to be able to quantify their performance. This isn’t easy, but I’ve made a video explaining the ways we analyse models in as simple terms as possible. I hope it helps you gain intuition behind what different performance metrics show.
This week I did a webinar and Q+A for Cambridge medical students interested in machine learning. They’re offering their services to people with discrete projects that need doing (full description at bottom of the email).
Have a great week!
👟 Know just enough for the next step
Our education system encourages us to “learn lots about X” before actually doing anything with the knowledge. We’ll spend a term in math’s class learning about differential equations. Then, maybe, we’ll study a subject (such as machine learning) where we see some real-world applications of that theory.
This creates the following habit in learning:
Decide to learn about X
Spend lots of time reading theory about X
Think about some projects that apply that theory
I’m a big proponent of the reverse approach:
Think of a project
Figure out what you need to know for the first step
Go and learn it
Repeat steps 2 and 3 for the next step
Of course, there is some benefit in having a basic insight into the area of the desired project, so that we can think of a good project idea. But that comes with experience.
📈 What do you gain?
This approach is way more efficient. We’re not spending time learning things we’ll never use. Plus, we’ll remember much more of what we learn if we use it. And, if we’re lucky, we may actually make something useful.
This also makes it much easier for us to get started. We’re not overburdened by what we need to know first. We can just start.
From experience, it can be uncomfortable at times. There’s an urge to stop and just read more theory. But I think this zone of discomfort is where the real learning takes place.
This is one reason I love hackathons. You’re forced to just start making something. You think of the project first, then learn what you need to make it. It can be uncomfortable but you push through. And you just have fun with it.
This week’s links:
(1) On not striving (a blog)
I enjoyed this blog, written by a software developer who works at Substack (the service I use to host this email!), which makes a case for actively not trying to strive and think deeply about thing.
(2) “Rationality and Life at the Margins” (Tyler Cowen on Tim Ferriss’ podcast)
I came across Tyler Cowen and marginal revolution recently. He’s an economics professor who has blogged every day for >15 years.
He’s quite a quirky interview guest, and I enjoyed many of the ideas from this podcast. In particular: (1) travel and understanding different cultures being the best form of intellectual expansion, (2) that we should read at least one tough fiction book and (3) the benefits of watching Spanish news.
(3) Why some people stay poor (an academic article)
I stumbled across this interesting paper, which describes an experiment conducted in Bangladesh to understand alleviation of poverty. That found that giving money above a certain threshold enabled people to get out of poverty and break the cycle, making a case against the “equal opportunity view” of poverty.
This week’s video:
How to Assess Machine Learning Models: AUC, F1 and more (clearly explained)
For a machine learning model to have a positive impact in healthcare, it needs to both perform well and to have a positive impact once integrated into clinical workflows.
This video covers how we can quantify the performance of machine learning models - for use in healthcare or otherwise.
Call for projects
The Cambridge University Students' Clinical Research Society is funded by the Wellcome Trust to facilitate research collaborations between students and academics. Our medical students are willing to improve their data analysis, project organization and writing skills. If you are interested in reaching directly a talent pool with a diverse set of skills, please write a short (< 200 words) description that we can post on our social media group.
The description should include:
Project outline (clinical, biotech, public health etc.)
Skills preferred (e.g. Python, R, Photoshop)
Estimated time commitment
Expected outcome (e.g. potential paper co-authorship, remuneration)
Send this to firstname.lastname@example.org and one of our committee members will confirm once it has been publicised.
Enjoy this email?
Please click the heart below, and forward the email to a friend!
Hi! I’m Chris Lovejoy, a Junior Doctor and Data Scientist based in London.
I’m on a mission to improve healthcare through technology (particularly AI / machine learning), but along the way I want to share learnings that are relevant no matter your career choice or background.
In this weekly newsletter, I share my top thoughts and learnings from each week, as well as links to the best things on the internet that I come across.