|
|
|
|
|
|
by h3ctic
2454 days ago
|
|
|
Linear Algebra is indeed essential - but Calculus as well. If you're talking about AI, then backpropagation would be the first thing that comes to my mind where Calculus is needed.
On the other hand if you talk about data science, I would focus more on statistics and probability theory (again linear algebra and calculus are tools you will need) |
|
|
1. Usually you're not implementing training, you're most likely using a framework that does training for you while you set up NN structures and data flows.
2. If you are implementing specific training methods yourself, massaging loss functions, new regularization functions and various objectives, and you are explicitly implementing backpropagation, then you're not implementing differentiation yourself, you're most likely using some kind of autodiff library or framework that does that for you.
3. If you are actually doing something with ML that involves you doing calculus, congratulations, you're likely doing PhD research on the topic and yes, sure, you're expected to know calculus at that point. But it's not really necessary even at this stage, an intuitive understanding of the concepts is obviously useful, but much of actual applied ML academic research doesn't really use calculus directly, the methods obviously use calculus but it's used just like CUDA kernels are used - obviously necessary but as a very low level of abstraction that's handled by those who specialize in it and most others can just apply it because it just works and is good enough.
I mean, seriously, you can get good progress and published papers in computer vision or NLP or data analysis without ever needing to directly use calculus; most people doing that would obviously know calculus as it's a basic thing, but it's not really that relevant to the daily work, most likely none of the experiments you'll run will require you to solve any calculus problems.