seems like a lot of "cutting edge" academic work is

>>57205983
Engineers really don't really wanna worry about correctly releasing their memory. They just wanna do their equations.

>>57205983
This gif has given me PTSD

>>57205983
Yes because nobody uses R nor matlab.

>>57206151
yeah i just got it from google and didn't watch the whole thing before posting, sorry

>>57205983
>signal processing/machine learning
Signal processing is a low level operation

>>57205983
>is this impression wrong?
nope

>>57206198
>>signal processing/machine learning
using matlab SP is pretty high level

Nothing has changed in programming in 30 years.

>>57206333
A low level operation used in high level language.

>>57205983
>It was after this chapter of his life Billy realized he could fellate his own penis.

>>57206198
>>57206333
Signal processing is not inherently low or high level. Real-time signal processing performs better in low-level environment. However the research on signal processing is more conveniently done using higher-level tools, unless the research aim itself is to optimize real-time SP algorithms.

Research code is meant to run a few times for you to write your paper and then be archived.

C++ will definitely give you a better solution, but improvements on code that doesn't run don't matter.

>>57206591
>machine learning
>code that doesn't run
eh?

>>57206610
Don't respond to the inbred (i.e. 99% of neo/g/) please.

>>57206610

In an academic setting, yeah. "Real-worlds" services will definitely consider C++ or Java.

Looking at major deep learning toolkits, for example, only Theano is not written at least in part in a low level language. Everyone wants to offer Python bindings though. For researchers.

>>57205983
Depends on the type of “cutting-edge” academic work you mean.

You aren't gonna be running python, matlab or R on a supercomputer cluster, for example. You also aren't going to be using them to explore axiomatic type theory. Nor are you going to be programming microcontrollers in them.

>>57206761
tensorflow is mostly written in python and the C subset is only usable to run graphs that were created with the python code.
Torch is lua.
Nobody uses caffe.
You are full of shit.

>>57206780
>You aren't gonna be running python, matlab or R on a supercomputer cluster, for example.
My sides shot out so fast they blew through saturn. Saturn's shards are now orbitting my sides.

>>57205983
Did he died?

>>57206165
>nobody uses matlab
Yeah, that's why MathWorks just printed $800 million in revenue last year, right?

Dumbshit.

>>57206859
He's right though, most of that shit is done in compiled languages like FORTRAN.

>>57206862
Yes.

>>57206165
>>57206925
physics departments control all kinds of shit with matlab, including MRI machines

>>57206843

Torch also uses C.

Fucking everyone uses caffe. Read any computer vision paper that uses deep learning, it will give you caffe binary models. Everyone tries to be compatible with caffe. OpenCV can only import caffe. tiny-dnn can only import caffe.

The Microsoft framework is C++.

The Yahoo framework is C++.

The Intel framework is C++.

>>57206198
Without more context this statement is wrong

>>57207045
Theano, TensorFlow and Keras on top. Comfiest setup for deep learning.
And all Python.

>>57206944
this

If you're lucky and your shit is CPU-based you're going to be using C++, C, FORTRAN etc. with libraries like MPI and OpenMP to help parallelization and message passing.

But for GPU-based clusters you're going to be writing CUDA or OpenCL

>>57207181

What you're describing is just Keras, it just runs on top of Theano or Tensorflow (which were already mentioned).

Don't get me wrong, Keras is comfy af, and the guys are porting popular caffe models. But it's still mostly in the realm of research.Would you build a large web service on top of it? Run it on embedded devices?

>>57206944
t. /x/

>>57207045
>it will give you caffe binary models.
>Everyone tries to be compatible with caffe.
>tiny-dnn
>The Microsoft framework
>The Yahoo framework
>The Intel framework
My sides just keep on splitting! At this rate, they'll go subatomic in 10 seconds!
>>>/x/

>>57207181
Keras and tensorflow are both outright unusable. Lasagne, torch and theano are the only usable options (lasagne being a higher-level option on top of theano and torch handling everything on its own). For example, tensorflow is literally so shit, it's 3x slower than theano for rnn, while also being an order of magnitude slower to compile an rnn graph, despite the fact that the only thing it does better than theano is compiling faster - but that only applies to shit where the compile time is already trivial.
Not to mention it's missing a metric fuckton of ops and has more bugs than google has shills.

>>57207511
Literally anything of note in Chemistry or Physics in done in FORTRAN or C++, or are just shitposting

For anyone wondering why there are no AI or ML threads on /g/, the answer is right here:
>>57207434
>>57207315
>>57207045
>>57206761
>>57206591

Anyone who knows anything about ML/AI will instantly leave and hide these threads when that's the kind of posts these threads generate. This in turn leaves the /x/tards to wallow in their blissful ignorance, giving them nothing to discuss. This causes these threads to die after 20 posts.

>>57207747
Nice strawman. Have you thought it up all on your own?

>>57206165
academics 100% use R. I missed out on a great work opportunity because of it.

Define "academic work". I'm a uni student studying Computer Engineering and we have a lot of genomic biologists using our supercomputer for sequencing shit in C and C++. When they have as much shit to calculate as they do (like 2TB of input data), you need a fast language or it'll take too long.

>>57207946
physics departments use matlab. comp sci often uses matlab when it comes to very mathematical computing like optimisation problems. used for DSP as well. as i said in OP they use C when speed is needed, just like in the example you gave with genes

>>57207771
So you are just maymaying then ?

>>57206108
You're confusing scientists with engineers

>>57208648
Scientists sure as fuck dont wanna do anything but their work, as use whatever tool is the most simple for them to learn.

you can get straight on with doing interesting things if you use high level languages. it's kinda like the difference between an architect and an engineer

>>57207754
Unlike you, I actually have access to a TOP500 cluster. If you're writing your own software, you're basically using OpenMP/MPI with your choice of C, C++, FORTRAN etc. (GCC and ICC/IFORT etc.)

There's some preinstalled crap like ANSYS, OpenFOAM, FASTA, QuTiP, MATLAB and what have you not for specific purposes, but I can guarantee you that none of these are written in fucking python or whatever the fuck meme skript language you people think is used for HPC

>>57209326
Take your garbage to >>>/x/

>>57206953
Source?