4 ms·
When I see work like these it get me the impression that ML hype is way too real. The goodness of the result of a machine learning model like this one should b
by siscia 7y ago
When I see work like these it get me the impression that ML hype is way too real.
The goodness of the result of a machine learning model like this one should be compared with the goodness of a simple "standard" model like linear regression.
Yeah it is kinda cool that we can use 10 lines of TF to spin up huge computation, but I guess that a simple linear regression would have provide results that are at least similar to the one of the neural network.
- fock 7y agothat's exactly the thing I'm seeing in my field (computational materials science). Basically a simple regression model (with very simple features) brings you 90% there, still people compete on publishing (on ONE shitty benchmark dataset) ever better results – the most-cited people are using KRR, e.g. each fit uses 2TB of RAM and "days" of CPU (features of length O(1000) and 100000 samples). The sample data is probably a 30s calculation (and still only a rough estimation). Sometimes wants you to question science, but hey, writing proposals with "ML" in it gives at least a chance on that grant...
- 1000units 7y ago--- guy who writes "O(1000)" telling you how ML research is flawed
- KaoruAoiShiho 7y agoI haven't thought as much about it as you have probably, but ok you get 90% of the way there, but now what? How do you get that last 10%? It would be a huge amount of work right?
- hef19898 7y agoIsn't that usually the case? That the first 90% are as difficult to achieve as the next 5%, which are again as hard as the next 3%?
- KaoruAoiShiho 7y agoBasically the argument is that ML would be able to get there easier.
- fock 7y agoyou get the last 10% by running a calculation for that (which itself is not necessarily accurate...). The whole situation is a little bit like Plato's cave... And also from what I've seen so far, those models either are 90% there or they are overfitting like hell.
- nilkn 7y agoThis is great if you only need 90%. But often that extra 10% is the difference between a state of the art commercially viable product and vaporware that nobody will pay for. It’s sort of like production software in general. Sometimes the core product is pretty easy to prototype... but serving it to all your users with very high reliability and uptime is not so easy, and that’s what actually gets them to pull out the credit card.
- sacado2 7y ago> Yeah it is kinda cool that we can use 10 lines of TF to spin up huge computation, but I guess that a simple linear regression would have provide results that are at least similar to the one of the neural network. Plus, LR is not a black box, so it both brings you a class and a reason why that class was chosen, which is a very desirable property in many problems.
- saystupidthings 7y agoThere are many ways to get around the BB problem. Calculating activations and uncertainty is enough to deduce reason, and throw out risky choices that should be human reviewed.
- ruborcalor 7y agoYeah i'd agree that ML hype is too real I guess I'm part of the problem It's more fun to use a neural net :) but after many similar comments I plan on implementing a simpler approach and seeing how it compares