This coming week is the last full week of the first term in my machine learning master’s degree.
So I thought it now would be a good time as any to reflect on the main things I’ve learnt. (Besides the specific course content, which is somewhat niche.)
1. Tailoring study techniques to the type of learning
Needless to say, studying machine learning is pretty different to studying medicine.
By the end of my six years studying medicine, I felt that I had my study approach nailed down for efficiency and effectiveness.
I had my spaced repetition-based system
in full flow, I had my mental frameworks ready for the insertion of new content and I was very happy with my self-perceived work-to-result ratio.
I knew the content of the machine learning degree was going to be very different; that the challenge was going to be more about understanding rather than the volume of knowledge to obtain (in medicine). But I underestimated the extent to which a different approach would be required.
I feel I’m just starting to get into the swing of things, having iteratively updated my approach throughout the first term. I don’t want to go into the detail about the different approaches here (although I will share one below), but am planning to write about this in future.
Regardless, I think there’s huge value in considering our approaches to learning, and have linked some fantastic videos exploring these at the bottom of this email.
2. Breaking down concept overwhelm
One recurring experience from this first term has been getting lost while being presented with a new concept, battling to catch up with the chain of logic and by the end thinking “what the hell is going on here?”
This wasn’t something I experienced in medicine which, although challenging in its own right, was rarely overly challenging in the conceptual sense.
When a new mathematical concept is presented in a textbook or a lecture, there is always a level of assumed knowledge. It would be excessively time-consuming to repeatedly build up first principles.
However, without a formal background in mathematics, the level of assumed knowledge has often been above my level.
As a consequence, when reviewing these concepts, I have found it easy to become overwhelmed. No matter how much I re-read the explanation, it wouldn’t make sense.
The solution that I found for dealing with this instinctive overwhelm is to visualise my lack of understanding as a bridge from ‘not-understanding’ to 'understanding’. However, the bridge has gaps and poor foundations, that must be filled to 'bridge the gap’.
I would first try to highlight the specific gaps in the knowledge that prevented me from understanding the concept. It could be something as simple as not being familiar with some mathematical notation, or something more complicated such as a rule of matrix manipulation.
I would then spend time studying the areas highlighted. If this brought up other gaps of understanding, I would keep on following those down. The idea was once I have filled all the gaps, built the foundational understanding, that I could keep stepping back up until eventually the full bridge has been created, and I understand the concept.
This may sound really basic, and I’m not certain I’ve explained it well, but this process of breaking things down was extremely helpful every time I felt this sense of 'concept overwhelm’, which was fairly often.
3. Not cheating myself on sleep
I’m someone who likes to take on a lot of side projects alongside my main work, and as a result try to maximise progress that I’m able to make each day.
However, sometimes this eats in to the time I should be sleeping.
In the adrenaline and excitement, I find I can often, with the help of a little caffeine, push on for several days on less-than-optimal sleep.
However, without fail, it always catches up with me; sleepiness starts to affect my motivation and the quality of my work, and then any progress gains in those initial days is counter-acted.
I’ve been reading a lot recently about the role and importance of sleep, and have just finished Matthew Walker’s fantastic book “Why We Sleep”. This, and other sources, have convinced me that this is something to prioritise going forward.