Building portfolio for AI/ML

[fast vessel]

Does anyone have recommendations for how to build up a portfolio in ML/AI to apply for jobs? I have a masters degree in the field (computer vision, medical imaging, etc.), but since graduating, I've mostly been working on software dev. Now I want to get back into what I studied, and while I have awareness of the theory, I don't have quite as much practice.

I know there's popular environments like PyTorch, but my main dilemma is figuring out a "starting point" to work on. For personal reasons, I'd prefer to avoid just working on models that have to be trained on large external datasets. I thought of making some kind of small utility that can learn from and emulate a local simulation, inspired by neural networks that were applied to particle and cloth simulations that I've seen some years ago ( https://www.youtube.com/watch?v=RXDaJTurAFU https://www.youtube.com/watch?v=iOWamCtnwTc )

[steel oak]

Bro this is a computer graphics discord

[weak raven]

Scroll the pytorch docs

It's a gold mine, there are tutorials for litteraly anything

And if you want to merge ai and graphics, just pick a ML denoiser lmao

That's a hot topic atm

And if you want some horsepower, just use google colab

If you want a good dataset to start training, use the BMFR one

Ping if you want the link

But the thing is: just google "<ml stuff> pytorch"

And with a lot of patience you can even port its layers to glsl/hlsl and see it running in real time (well most of the time, pick a tiny one)

The hello world in ML (deep learning) is a classifier, you could start with that (assuming you want to jump to CNNs)

There are some wild stuff with cellular automata (applied to mesh and texture generation), you should check that out. Cellular automatas are simple by design btw.

[fast vessel]

There are ML applications related to graphics. Just ask NVIDIA

[fast vessel]

That's a new one. How do CA and ML techniques relate to each other?

[weak raven]

I mean that's a weird question, cause ML is a bit too broad as a term

It contains DL, CA, clustering, decision trees, whatever

Anything that "learns" in a way or in another

My answer to your original post was a bit targeting ML

[fast vessel]

Sorry, yeah, what I mean is, I've never heard of cellular automata being used as a form of machine learning. I know it's theoretically possible, e.g Game of Life can do arbitrary computations, but is it also practical?

[red anchor]

If you're interested in graphics + ML, then check out radiance fields. Build your own NeRF or splatting implementation. There's even a discord for it here: https://discord.gg/fDv4Qwfh

[steel oak]

Sry for initial response, Gaussian splats are a very cool intersection of graphics and ML I’d look into if you want to fuse the disciplines. definitely possible to scratch both the ML and graphics itch at the same time if that’s what you’re going for. For example I work in graphics programming for robotics applications / simulations for training real world AI and mixing the two definitely has been a joy.