2
$\begingroup$

I'am finishing my undergraduate degree in computer science and despite having had to take some math classes my math ability is still pretty poor. I struggled a lot with it which I believe was due to missing some pieces of knowledge that I needed to know and not seeing the big picture. Now i'am studying neural networks and turns out they require quite some math(eg: the backpropagation algorithm uses the chain rule). Some people say you don't need the math but I don't think I will be able to to understand neural networks completely without it. And even if I could get away with I would probably still need the math later on.

I have about a month that i can dedicate fully to this quest and I'm determined to learn linear algebra and multivariate calculus in this time frame. I will start with linear algebra (doesn't depend on calculus, right?) and I plan on learning from MIT 18.06 video lectures as well as doing the assignments. They have solutions so I should be able to easily track my progress. I also found this course from Berkeley Math 110. Linear Algebra. It doesn't have video lectures but it has more assignments with solutions so I can practice even more. As for textbooks MIT uses "Introduction to Linear Algebra, Fourth Edition, Gilbert Strang" and Berkeley uses "Linear Algebra by S.H. Friedberg,A.L. Insel and L.E. Spence,Fourth Edition". People here seem to say good things about them.

I'am starting this journey tomorrow. In the mean time I'd like to get some advice. Do you have any advice in order to make this process smoother? Are there any other resources I should know about?

$\endgroup$
  • 5
    $\begingroup$ Best advice I would give is to take more time than a week. Sometimes you need to sleep on a topic to understand it better. "Rome wasn't built in a day." That's not to say that your attempts will be in vain, just that you will probably have a much easier time if you stretch it out longer. At the very least (I'd still prefer taking a few months), instead of one week on linear algebra followed by two weeks on calculus, try going back and forth between the two over that time frame. $\endgroup$ – JMoravitz Jun 10 '17 at 20:52
  • $\begingroup$ If you have few time, it's not a good idea to switch between 2 (big) books. I only know Strang which is excellent ; hint: cover only the first chapters. Stop before quadratic forms that deserve a few more weeks... $\endgroup$ – Jean Marie Jun 10 '17 at 21:02
  • $\begingroup$ My advice would be to ultimately have the goal of understanding various applications of linear algebra (which shouldn't be hard to find in intro textbooks/courses) and understanding how linear algebra is a framework for solving those problems. E.g. compare and contrast the use of linear algebra in solving recurrence relations, hamming code stuff, least squares stuff... If you want a deeper understanding. $\endgroup$ – mathematician Jun 10 '17 at 21:06
  • 1
    $\begingroup$ I would cry if I had to learn linear algebra in a week time. It really took some time for me to grasp the concepts, especially the matrix/linear transformating correlation. I strongly suggest to take more time for that. $\endgroup$ – user370967 Jun 10 '17 at 21:15
  • $\begingroup$ I learned introductory linear algebra over the course of about 2 months and that was studying it for at least 5 or 6 hours every day. So what I'm saying is ... Good luck, mate. ;-) $\endgroup$ – user137731 Jun 10 '17 at 21:16
2
$\begingroup$

I personally study Neural Networks. You require a solid mathematical background to grasp the topics, especially derivations of different learning rules.

As an introductory text, I suggest "Neural Network Design" by Hagan, Demuth, and Beale. The textbook is written in a manner to first introduce the mathematical concepts, and then apply it to neural network scenarios. They do a good job of reviewing concepts in linear algebra, multivariate calculus, and optimization.

Additionally, if you want to succeed at anything I think you need to take your time. Don't rush :)

$\endgroup$
  • $\begingroup$ Yes "crash courses" may be lead to ... a crash. $\endgroup$ – Jean Marie Jun 10 '17 at 22:25
  • $\begingroup$ @Joshua_T thanks for the book recommendation, i will take a look at it. $\endgroup$ – plx Jun 11 '17 at 5:32
1
$\begingroup$

Another way you can learn apart from doing Strang and co's lectures is to check this book out. I'm going through this at the moment: Coding the Matrix. Look it up on Amazon. You can get it as a kindle edition which means you don't have to wait for it to arrive. It teaches linear algebra from a more practical and applications perspective and also teaches you python at the same time. It is based on a linear algebra course at Brown University. The guy who wrote it seems to more than just a good mathematician, but good at writing content in a way that maximise knowledge transfer.

$\endgroup$
0
$\begingroup$

Thanks for the advice guys. I realize now that the time I have is too short to get a good grasp of linear algebra and multivariate calculus. I also don't have much time to implement the neural network. What I'am going to do instead is study neural networks top down. If a mathematical concept is needed to understand some part of neural net I will study that math more in depth.

After this project is finished I will rethink what to do. I still want to learn linear algebra and calculus so I will try to find a way of doing that but without rushing.

$\endgroup$
0
$\begingroup$

https://www.lem.ma/books/VBS92YDYuscc5-lK/problems

this site contains more than enough questions and problems and has solutions. Hope this helps.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.