Plan to merge three similar packages

[yuehhua]

Since this project is mature, is there any plan to move to FluxML?

[foldfelis]

Yes, sure, is there anything I need to do before I transfer the project？

[yuehhua]

@DhairyaLGandhi Could you help this?

[foldfelis]

Hi @DhairyaLGandhi, sorry for the late reply. I don't know why I didn't recive the notifecation of the FluxML invitation.
I am currently discuss with @ChrisRackauckas about the future plan. Since both NeuralOperators.jl and OperatorLearning.jl provides solutions to the same problem, it might be a good idea to combine the efforts.

[ChrisRackauckas]

There's a third one too: https://github.com/CliMA/OperatorFlux.jl @bischtob @charleskawczynski and pulling in @pzimbrod . There's nothing wrong with having multiple repos of course, but I think finding the most useful ideas of each repo and folding them together to get top notch performance would be more useful in the end.

Note that one of the plans I have for where OperatorLearning can go is, given its small neural networks in many cases, we have a fully pre-cached ML library (right now called SimpleChains, being set public fairly soon) that is extremely fast for this small NN case by being fully non-allocating and direct usage of LoopVectorization. At least for the DeepONet parts that can be very useful.

I hope to get a GSoC going and some Julia Lab students going on this as well.

[bischtob]

I am down to work together on one repo @ChrisRackauckas ! Would be fantastic to collaborate.

[pzimbrod]

I would definitely chime in on the idea of combining our efforts. I think there still is enough functionality and optimizations yet to be implemented and I'd love to see that field grow further in Julia, so why not work on it together 🙂

[yuehhua]

@ChrisRackauckas I am also interesting in contributing this work. There is another work for multi-wavelet operator ongoing. My direction is bringing neural operator, GNN and geometric deep learning together. If there is anything I can help, just let me know.

[ChrisRackauckas]

Should we have an "internal conference" demoing the 3 packages to try and figure out the difference? The best time I can think of would be 6pm EST (midnight Germany, 7am Taiwan, 3pm west coast. I think that might be the only time where someone isn't in the dreaded 1am-5am range?) 15 minutes each, no slides or anything, just demo the package and figure out what to do? Monday the 21st would work for me. Otherwise I think we'll end up in a bit of deadlock.

[yuehhua]

The 21st, 6pm EST? It would be 22nd, 7am Taiwan. It works for me.

[foldfelis]

Works for me as well.

[pzimbrod]

That would be okay for me as well

[bischtob]

Monday 21st works for me.

[bischtob]

Where should we meet? Zoom?

[ChrisRackauckas]

Invites went out if I have your email. Otherwise just message me your email to get you on there.