Learning track for software dev

I am currently taking the Mathematics for Machine Learning and Data Science Specialization, and trying to line up a plan for further courses to learn ML.
…but…
If the Machine Learning Specialization mentions “don’t worry about the math”, and the Mathematics for Machine Learning and Data Science Specialization stance is “don’t worry about the ML” . . . then where do I go after I’m finished both?
Thanks,
T

1 Like

You will be slightly confused after completing the Math for ML course and the MLS courses. They teach similar materials but in significantly different ways.

The MLS courses are designed to feed directly into the Deep Learning Specialization.

Ok thanks! Looking forward to advancing along.

Is there any sense in taking both of the first two?

My opinion only. Others may not agree:

If you enjoy math for its own amusement, then M4ML is useful.

MLS has just enough math to understand why machine learning methods give useful results.

2 Likes

I agree with Tom. But it’s worth saying that the key math you need to understand MLS and DLS is linear algebra. You don’t need anything as advanced as eigenvalues and eigenvectors, but you need to be very comfortable with vectors and matrices and algebraic operations on them and operations like transpose. You need to know what a dot product is and how matrix multiplication works and have prior experience using those operations.

If you already have that, you can take MLS first and then DLS.

If you don’t have that level of knowledge of linear algebra, then you can try the first course of M4ML, but that’s really all you’d need to start MLS. The calculus and probability from M4ML C2 and C3 may give you some greater understanding, but MLS and DLS are designed not to really require calculus. Of course calculus is involved, but Prof Ng just gives you the formulas. If you want to know why the formulas are what they are, then it will help to know some calculus beforehand. But it’s not clear that M4ML C2 is really enough to be on solid footing if you actually want to derive all the formulas for back propagation in DLS. For that, you’d need the equivalent of the first two years of math courses that a math or physics major would take at a college level.

3 Likes

Let me just give an example. This is a slide from the MLS:

See? Andrew wrote down and told us “that is a vector” image, and “that is a dot product” image
.

Here image, you know a vector is a list of numbers which are the parameters of the model.

Here image, you see what a dot product represents - a bunch of multiplications and summations among the elements of the vectors in the dot product.

Certainly, Andrew won’t go into many examples of dot product, but if you were not familiar with these algebric concepts, then you might actually pause there and try to connect things up. For example:

You see - Andrew and the course team intentionally used different colors for the elements in w vector and x vector so that you can connect them easier. He would also go through these slowly but you always can control the pace yourself.

If you accept what Andrew tells you, then you can make notes and do whatever to strengthen your memory, and then you can move on. This means it is going to take extra time, because you might see new maths from time to time, and you might see them being used in new ways once in a while. However, in this way, you are assured to be on the track of learning ML and only necessary maths for learning ML, if that is your goal. Of course, if you insist to have some extra reading, you will also know exactly the name of the topic to search for, here, for example, dot product. This is important because you won’t get side-tracked.

As a beginner, I think this is a sufficiently good way, perhaps you wouldn’t be able to expect to become a math expert afterwards, but over time you should be more comfortable with seeing them in ML context. You might not expect to do the calculation quickly in your mind but the MLS will tell you how to do them with computer, in Python. You probably wouldn’t expect to teach maths better afterwards, with a lot more of insights I mean, but you could tell the others what maths are oftenly seen.

I think, in this way, you could really leave the decision for a Maths course later. During or after MLS, you can decide whether to dig deeper into the Maths, or you might think that, alright, that’s enough so far, and after the MLS, I want to try a more advanced deep learning course and program something out!

Cheers,
Raymond

1 Like