Hacker News Reader

Comment by mattlutze | Hacker News Reader

Well, a lot of / most of deep machine learning breakthroughs are coming directly from research scientists at Facebook, Google, NVIDIA, (other major tech companies) or research organizations associated with them or their executives.

It's difficult for academic labs to keep up, actually.

Whilst modern economics and the startup model makes it easier for companies to let startups do the actual idea testing (and then buy up the successful experiments), there's a ton of companies that are still driving entire industries with their corporate-funded research.

We should also probably note that a lot of the funding for some very successful academic labs comes from these large corporations as well.

View on HN

> It's difficult for academic labs to keep up, actually.

That's probably more an intrinsic problem with academia (lower wages, limited career opportunities, limited resources compared to companies).

And also an artifact of strategic choice of pursuit of these technologies.

Due to the resource and scaling requirements (as well as inherent skill needed), replication of efforts for expansion purposes is difficult and that's one reason (not the only reason) these sort of modeling approaches are being pursued.

A lot of technology choices are being chosen because of an inherent high barrier to entry, limiting competition risk.

There is a spectrum of the kinds of research one might do. My advisor always used to remind me that if a team of good engineers at a big corporation can do something, it's not a good research problem to tackle for a grad student in a uni lab. So we naturally tended to more speculative/fundamental problems, which seems like a good thing - instead of trying to compete with industry labs we augment them.

That sounds like a good thing, but realize that some day industry will build an application based on your ideas and patent the result. Patents are a monopoly on technology, and working for a university, probably the last thing you had in mind was that other people would monopolize your inventions so that their use would become limited.

University labs are generally good at dealing with patents. Researchers do generate patents where it makes sense, and the processes are set up such that the university gets some ownership with the option for the researchers to transfer the knowledge to companies e.g. student graduates and founds a startup with the patent as foundational IP. This is all intentional and it's one of the ways for the labs to do their job of disseminating their discoveries and transfer knowledge to industry.

Still, patents are a way to provide exclusive access to technology, to those who pay. That's not something I personally would like to see happening to my inventions if I'd work at a university. It's also not how universities used to operate, traditionally.

Further, I feel that if it's paid for by the public, it belongs to the public.

My advisor made the exact same point. Although we did receive funding from startups to work on adjacent problems.

This is kind of odd logic though. If a well funded team of experts who care about the utility of the results would focus on a problem, you shouldn't.

What makes you think the advisor's projects were genuinely more speculative or more fundamental vs just dead ends that nobody cared about? It's not like corp labs don't fund fundamental work. Google was funding quantum computers for a long time now.

This is how my advisor and I reshaped my recent masters thesis. I was thinking of an engineering problem, but he reminded me of some really cool ways we could do basic science. It ended up being really rewarding and still novel.

Reply on news.ycombinator.com