Thread 16834165

Is this book enough for a Masters in Machine Learning?

https://udlbook.github.io/udlbook/

>Is this book enough for a Masters in Machine Learning?
You want a masters degree for reading a book?

>>16834165 (OP)
>Masters in Machine Learning
since that would be a fake and gay degree mill program, sure

for an MSc in statistics or computer science, no.

>>16834191
I'm retarded
>>16834194
Okay, so what else would I need to read for a Masters in ML?

>>16834165 (OP)
> Is this book enough for a Masters in Machine Learning?

No, you'd also need to do some original research and write a thesis, and probably some coursework (i.e., a master's degree).

For what it's worth, Bishop's new Deep Learning book is far better on a technical level.

>>16834165 (OP)
Looks fine to me. Read some papers as well, maybe throw in https://d2l.ai/

You can also just take Georgia Tech's OMSCS for relatively cheap.

>>16834191
yes please

>>16834196
nothing, just enroll, pay the fees and use chatGPT to do the assignments they give you. since your instructor will also be using chatGPT to check your work it doesn't matter

>>16834199
>you'd also need to do some original research and write a thesis
unless you're some prodigy hypergenius you're not doing anything really original for your master's thesis even in real fields. and the prodigies who do that typically just get a fast tracked PhD at that point anyway.

a master's thesis is just a 50-100 page student paper that shows you know how to write a technical/research report.

Lol

You need to pay me in advance if you want me to give your institution a credit it doesn't deserve

Care to mention georeferenced porn generators?

To undergrads?

>>16834325
> unless you're some prodigy hypergenius you're not doing anything really original for your master's thesis even in real fields. and the prodigies who do that typically just get a fast tracked PhD at that point anyway.

Maybe I go to some weird outlier school (I don't think so, it's just a pretty decent state school in my engineering field), but every master's student I've ever worked with has done original research. Usually it is a marginal contribution (e.g., show that this previously published method can be used to solve this other problem), but it is original in that nobody else has done it. I'm pretty sure our department would not approve of a master's thesis which was just some bullshit regurgitation of a textbook or primarily a literature review.

>>16834165 (OP)
Just from skimming it, it seems kind of strange. It talks about the various commonly used activation functions before even using the word "matrix" almost 50 pages in... I don't know how you can accomplish that as an ML textbook. Unless you have some basic understanding of how matrix multiplication is used in a ANN, there's no intuitive reason to use nonlinear activation functions for each layer
Also goes into a lot of detail about loss functions before actually talking about optimization methods, which also seems strange to me, but maybe that makes pedagogical sense
>Masters in Machine Learning
No lol

>>16834383
you seem to have a wildly different definition of "original research" and "marginal contribution" from mine.

the average PhD thesis is a marginal contribution, with a bit of original research. yeah they'll make you show that you can sell your paper as somehow novel and original in a master's thesis but it really won't be in any meaningful sense. you can generate a just about infinite number of perfectly acceptable master's theses without ever pushing the field forward.

>>16834452
> you can generate a just about infinite number of perfectly acceptable master's theses without ever pushing the field forward.

You could say the exact same thing about dissertations. "Pushing the field forward," is not something that any particular researcher has control over. My most cited paper is one that I considered to be a middling contribution, because it reformulated an existing problem into one which then inspired a lot of alternative approaches.

My most ambitious paper was almost 30 pages of novel work which had the potential to open many new related works. It has one citation from outside of my general orbit, and it was in the literature review section of a paper from some random European university that had basically nothing to do with it.

Scientific literature doesn't move forward solely by papers which take years to write and require technical wizardry. It moves forward by a mixture of finding new ways to solve old problems and old ways to solve new problems.

>>16834324
ChatGPT creates the exercises (for the teacher), answers them (for the students), and grades its own answers (for the teacher again). It's surreal.

>>16834643
Extremely based. I can just run 2 local models and have them learn real analysis for me

>>16834587
>My most cited paper is one that I considered to be a middling contribution, because it reformulated an existing problem into one which then inspired a lot of alternative approaches.
>muh citations
if you've made a genuine contribution you will have thousands of citations. otherwise it's usually a matter of
>I need a source for this claim I know is true
>lemme grab something off semanticscholar real quick

In any case, my point stands. A master's thesis does not contain meaningful contributions to anything, and does not require the amount of work that would entail. The vast majority of master's theses are just fluff, and they get assessed on technical skill rather than impact.

>>16834587
>>16835165
And
>You could say the exact same thing about dissertations.
yes, exactly. A dissertation requires years of work and deliberate research, and even that rarely amounts to anything other than demonstrating that the particular individual behind the thesis is qualified to receive funding for research.

>>16834194
>masters in computer science
I cannot imagine anything more worthless

>>16835430
I received a mail today where my faculty is talking about a bachelor in AI till next year...

>>16835165
> In any case, my point stands. A master's thesis does not contain meaningful contributions to anything, and does not require the amount of work that would entail.

You have a very naive view about how scientific contributions work. You can clearly see this here and below:

> A dissertation requires years of work and deliberate research, and even that rarely amounts to anything other than demonstrating that the particular individual behind the thesis is qualified to receive funding for research.

Firstly, you seem to be holding yourself and others to a standard in the neighborhood of, "if it isn't in the running for a Nobel prize, it isn't a real contribution." That is a very stupid standard, and you would be eliminating the super majority of scientific developments based on this kind of thinking.

You would be much better served by getting your head out of the clouds and accepting scientific progress as it has always been "slowly, slowly and then all at once." For every 10,000+ citation game changer, there were hundreds of smaller contributions that were necessary to set the stage for that to even happen in the first place. There is nothing wrong doing honest work and planting trees you may not be around to enjoy the shade of.

>>16834165 (OP)
i think this book is decent, and a better read than the bishop DL book. it's a good way to get some intuition for common methods and is nicely illustrated.
t. ML PhD student