ML papers by Western universities barely touch on the problems that practitioners face.
The only papers I see that are routinely useful have half the authors having a .in or .cn email at the end with the rest having Indian and Chinese names in US institutions.
The only western papers which aren't extended advertisements for their company are from people who are making something for themselves.
For example the best paper on image classification I've ever seen was posted on a private discord and was about better labeling the parts of a vagina as part of a stable diffusion training pipeline.
I used the methods without change and got better than state of the art for document segmentation.
Certainly, some countries have a more engineering-focused academic style. Western academia has always been more about advancing knowledge, which IMO is academia's mission.
> Western academia has always been more about advancing knowledge, which IMO is academia's mission.
You can advance knowledge in ways that are aligned with the nation's strategic needs. That would imply the career path of researchers would be oriented towards industry instead of pie in the sky projects.
I fail to see why universities should align on their country's strategic interests. Universities are not political nor military entities. Additionally, pie in the sky projects is what's needed to advance science, which is very distinct from advancing technology (industry).
> I fail to see why universities should align on their country's strategic interests.
In the UK, the industrial and military research labs have been closed, there is nothing being done except in universities. The university projects are really specific and don't join up in any way. There is no equivalent of DARPA guiding
research proposals to be useful but Governments still think that the universites will define the next generation of industrial products.
Source: my own area of trying to use computers for Materials Science.
Yeah. Practical working implementations of the latest in ML image classification technologies are a rapidly changing incremental improvement problem that industry is already all over, so not really surprising that this isn't a major focus for US university research: if PhDs or even potential PhDs want to do that they can get a much higher rate of pay at a private company.
All the foundational work that has lead up to the current practices in ML was done at universities. It's not like Google invented the transformer from scratch completely over night.
The only papers I see that are routinely useful have half the authors having a .in or .cn email at the end with the rest having Indian and Chinese names in US institutions.
The only western papers which aren't extended advertisements for their company are from people who are making something for themselves.
For example the best paper on image classification I've ever seen was posted on a private discord and was about better labeling the parts of a vagina as part of a stable diffusion training pipeline.
I used the methods without change and got better than state of the art for document segmentation.