3 ms·
> "you only need high school knowledge" means nothing to most people. I think people intend to use it to tell people the barrier is low. But by trivializing th
by godelski 2y ago
> "you only need high school knowledge" means nothing to most people.
I think people intend to use it to tell people the barrier is low. But by trivializing the difficulties of calculus (may be easy now, but was it before you learned it?), you place that barrier higher than it was before. The result is the opposite of the intent.
I'll even state it now, as someone who highly advocates for learning math:
You don't even need calculus to build good models. At most, a rudimentary understanding of algebra, but I'm not sure even that. A little programming skill, which can freely and easily be obtained, is all that is necessary to begin. So if you can read and can motivate yourself, you can build good and useful models. It might just take longer if you don't have these yet.
With that said, be cautious that you fall victim to your success. The barrier to entry may be (very) low, but it is a long way to the top. So don't ignore the fundamentals and use your excitement and success to motivate yourself through the boring and hard parts. Unfortunately, there's a steep curve to reap the rewards of your math knowledge (in ML. You'll reap rewards even in daily life much sooner!). But it is well worth it. ML is only a magical black box because you have not achieved this yet (this does not mean ML becomes a white box). Not knowing what you don't know makes it hard to progress. But I promise math will help illuminate things (e.g. understanding when and where CNNs vs transformers should be used inside architectures; how many parameters you need in hidden layers; how to make your models robust; why they fail; how to identify where they will fail before it happens; and much more. These are enormously helpful, and more if you wish to build products and not just research papers or blogs. If models are black boxes due to the composition of easily and well understood functions, I think you can probably guess how small and subtle changes can have large effects on performance. You'll at least learn a bit about this concept (chaos) in differential equations).