Machine learning / Deep learning / Data science

Tell me user, what have you been working on, any cool models?

Attached: 1648848392526.jpg (1920x1280, 209.11K)

Bumping for self interest

I have been donating CPU and GPU cycles to MLC@home, to help aid the effort in fully understanding how/why our ML and NN algorithms actually work.

Recently I have messed with autoencoders, convolutional autoencoders, pretrained CNN models for feature extraction, and movenet for human pose estimation. It was fun to experiment with, but the unsupervised results were poor and I gave up.
It's not the first time I play with neural networks and get nowhere. I think that they are only good for specific things like supervised image models trained on huge datasets.

total newfag to ai, what is THE way to make an image sorter?

I'm trying to figure out how people are meant to run ML models that don't fit in memory.
All of the biggest (literally) meme models are cluster-sized and take like 20 GPUs
But what about those of us who only have 4GB VRAM?

Nothing works. Sort by name or metadata.

uselss because what I am downloading isn't tagged with metadata and is instead very specific and could be discriminated against.

Bump

the only serious way to manage your CP is to tag it yourself. You can build a model based on tags you've done yourself, but ultimately you're doing a lot of the monkey work yourself

I already have enough tagged to train, I just need the software to train it.

bumo

MLC@home is a grift, user. All the 'explanability' shit, in the first place, has been tainted by literal trannies complaining that apes are being tagged as gorillas and that's nod fair ;_; so we need to understand the models so we can censor them better.

You aren't.
Most those models are not worth running anyway, and advances come from more careful model design and data selection rather than gpu farm flex.

Just run it through sklearn primitives like randomforest, gradient boosting, adaboost or svm's.

Not quite. You can also use unsupervised algorithms to greatly improve your workflow: choose just a few suitable pictures, run some autoencoder or other, gather all images within a certain distance of the few images you selected. This generally works really well to at least do a first-pass sort.

I see you have no proof.
Why should the how and why of ML and NN remain a mystery?

Fuck off, grifter shill.

Please prove that I am a grifter shill.
I think it is important to understand the how/why so that we can produce better models in the future. You've postulated that that is the wrong way to do things, but seem to be unable to explain why.

First I've heard about it.
Sounds like a waste of effort.
ML models don't need to be explained, they need to do the job.
So what if a bunch of undergrads can't handle the math, or if your model encodes latent information about society that it wasn't supposed to?
It's not your problem.
Make a different model to do something else.

It is good to want to explain ML models: this way you can get a meh estimate to a very complicated problem, extract the patterns the ML model found, and use that to guide an analytical solution that could achieve much better results. Or simply to be much faster. You can also use it to understand the underlying nature of the data: what part of the input interacts with what part of the output? Especially: what parts of both are completely irrelevant/random/etc.?

The problem is that the entire area was taken over by trannies so it's not what any of this is about.

Guys what the actual difference between machine learning engineer and data scientist?

I want to transition from software developer to either machine learning engineer or data scientist position.

Google throw a lot of info and so many people saying "engineer" is better, as usual when interviewer hearing "engineer" they have a boner I don't know why.

Anyway I wanted to hear the answer from the people in the field.

Also how is the pay?
Is your daily life a brighter than say a soulless SWE code monkey?
If you where to choose between SWE, machine learning engineer and data scientist, which one would you choose?