We currently have a client in the insurance industry that will need MLE work specially in the mlops side of things.

Send me an inbox if it's of interest.

I'm in the same boat. Years in this ML business full-time. When I get out of work, I don't want to touch anything tech related, not even with a stick.

I found this one under the keyword "embedding fusion" in llm:

https://arxiv.org/abs/2101.12294

It provides overview of many methods.

And as other said anything on multimodal fusion transformers.

In the UK there are at least 3 types of PhDs: 1) PhD by thesis, 2) PhD by publication and 3) profesional PhD.

Options 2 and 3 may be compatible to what you want. It really depends on your advisor, the source of funding, and your motivation. There are many reasons to pursue a phD that doesn require an academic motivation and you are entitled to your own reasons. I've met people that have done PhDs while also having a Job but those are the exception rather than the norm and challenging to say the least.

Skimmed through a book about "fast and frugal decision trees" found it interesting. I don't think it is something new per se, but I found the concept useful in terms of data efficiency and explainability.

Check out: repeated-measures datasets, and hierarchical datasets. Might be useful.

What if you export vgg16 from tensorflow to onnx and load it on pytorch and compare that (and the opposite as well) maybe more comparative.