Hello,
First thanks for offering this course, material and teaching seem amazing!
(Also, hope you are recovering well).
I would like to know how long this course can be accessed? It is free for now, but do you plan to change it to something else later?
Can I suscribe now and download video and material for home use (not sharing these but instead your website address), or are there parts of the material that won’t be accessible if I waited say, 3 months, as I have to finish other MOOCs I have in progress to get a certificate.
I don’t have 7 weeks for now to make it but I am deeply interested, as I like this approach.
I wish there was a similar course for other ML methods, as there is no free lunch algorithm;)
I was aware of this mooc starting, but actually, I feel like it’s a missed opportunity, as I read now that you are starting a phase 2 that I won’t be able to attend (I will soon have more free time, but won’t be able to finish part 1 in time to apply for attending part 2 as an online student.
Do you plan to do other sessions? Do you plan to do live sessions in Europe?
If you were to make a math course for machine learning in such form, I would go all “shut up and take my money!”, too.
as for other suggestions,
Maybe it is not at all technical, but I would like to learn about the importance of biases to the models.
More specifically, the biases related to the data we use to train it, and the consequences in real world. It is not just about missing generalization, but also about real consequences for real users or “audience” (like when you build a system to help deciding if which rate you would apply in your insurance for a given person).
I think that NN, as they are such sensitive, can be even more biased than regular methods.
I would like to know about how constraining problems in NN/Deep learning like telling it that solution must be in some range, some sign, etc, apart of specifically designing your cost function or kernel for this.
I am also wondering about techniques allowing to integrate data with some relations between values;
usually on your input layer, you just feed a number. What if for some reason, you would have data tuples, which could be interesting to process as such. I find the limitation of having to use only arrays of numbers quite bothersome for some ideas I had for solving past problems.
Would you need a translator for other languages (I’m french and it’s one of my jobs apart of science)?
Thanks for your help.
Best regards and keep up the good work!